Zoning Data in the U.S.: Sources, Coverage, and Reading Codes for Development Screening

Every commercial real estate screen eventually asks the same question: what can this parcel legally hold? The answer sits in two places that are not the same place. One is a map, a polygon layer that assigns a district to a piece of ground. The other is a legal text, an ordinance adopted by a city council or a county commission, amended on its own schedule, and written to be read by a planner rather than parsed by a machine. The map is published widely and cheaply. The text is not. Almost every claim you will read about zoning data coverage is a claim about the map.

The gap between the two is measurable, and it is larger than the marketing suggests. California's statewide zoning compilation, assembled by the Governor's Office of Land Use and Climate Innovation from 535 of the state's 539 jurisdictions, publishes 568,741 polygons across two files. Its northern file carries 12,091 distinct local district codes and its southern file 8,886, and both collapse into the same 14 standardized classes. Neither file carries a single minimum lot size, setback, height limit, floor area ratio or parking requirement (California Governor's Office of Land Use and Climate Innovation, California Statewide Zoning, queried August 24, 2026). Oregon's statewide layer is cleaner and older: 114,823 polygons, 56 standardized statewide classes, and the local district code populated on 653 of them, which is 0.6% (Oregon Department of Land Conservation and Development, statewide Zoning feature service, queried August 24, 2026).

That is the counterintuitive finding this piece is built on. The zoning map is the cheap part of zoning data, and the parameters that actually determine what a parcel can hold live in unstructured text that no national program collects. In San Francisco's own assessor roll, only 8,941 of 211,547 parcel records for roll year 2025 sit in a district whose tabulated floor area ratio is not disapplied by a subsection of the same code (San Francisco Office of the Assessor-Recorder, Assessor Historical Secured Property Tax Rolls, queried August 24, 2026; San Francisco Planning Code Table 124 and Section 124(b), 2026). In Chicago, 1,446 of the 1,528 zoning class values in the city's current layer are individually adopted Planned Developments whose rules exist only in a bespoke ordinance (City of Chicago, Boundaries, Zoning Districts, current, queried August 24, 2026).

What follows is a working method rather than a directory. It covers why zoning resists standardization, who publishes zoning data and in what form, what a published layer actually contains, why an assessor use code is not a zoning district, why the parcel is the wrong primary key, the short list of parameters a development screen genuinely needs, and the verification chain that turns a polygon into a defensible statement. It belongs to the wider argument about the public-data stack for commercial real estate analysis, and it sits directly upstream of estimating buildable area from a parcel, which is where these parameters get turned into a number.

Zoning is a legal text and a map, adopted tens of thousands of times

Parcel data has a natural producer. One office per county maintains an assessment roll, one office maintains the polygons, and while the quality varies enormously the structure repeats. Zoning has no such regularity, because zoning is not a record of what exists. It is an exercise of the police power delegated by a state to its local governments, and the delegation is not uniform.

Start with the denominator. The 2022 Census of Governments counted 90,837 local governments in the United States, of which 38,736 are general purpose: 3,031 county governments, 19,491 municipal governments and 16,214 township governments, with 39,555 special districts and 12,546 independent school districts accounting for the rest (U.S. Census Bureau, 2022 Census of Governments, Organization component, Table 2, released 2023). Municipalities and townships together are 92.2% of the general purpose total. Township governments exist in exactly twenty states, and in New England and parts of the Mid-Atlantic the zoning authority is the town rather than the county, which is why a national zoning compilation is not a 3,000 county problem.

Not every one of those governments zones. The National Zoning Atlas puts the number of zoning jurisdictions in the United States at more than 33,000 (National Zoning Atlas, read August 24, 2026), which is the only published national count of zoning jurisdictions this research found in a public source, and it is a nonprofit's figure rather than a federal statistic. Even taken at face value it means that the object being mapped is roughly ten times more numerous than the counties that produce parcel records, and each instance is an independently amendable legal document.

Three structural facts follow from that, and each one breaks a habit imported from parcel work.

The first is that there is no federal mandate and no national standard schema in force. The federal government has never enacted zoning; the model that most states copied is the Standard State Zoning Enabling Act of the 1920s, and what each state enacted from it differs. There is no federal zoning dataset, and the statewide layers that do exist are compilations of local maps, assembled by a state agency or a university that asked each jurisdiction for its file. The federal content standard for parcel data, the Geographic Information Framework Data Content Standard, Part 1: Cadastral (FGDC-STD-014.1-2008, May 2008), does not address land use at all, a point examined in detail in the companion work on lot size, coverage and assembly patterns.

The second is that the parcel is not the primary key. Zoning districts are drawn as areas, and an area boundary is under no obligation to follow a lot line. A single tax lot can sit in two districts, carry a commercial overlay on part of its frontage, and fall inside a special purpose district that rewrites the base rules. New York City's own data dictionary states the consequence plainly: where a tax lot is divided by a zoning boundary line, the primary zoning district field records whichever district occupies the greatest percentage of the lot's area (New York City Department of City Planning, PLUTO field documentation published with the PLUTO 26v1 dataset, read August 24, 2026, and the PLUTO Data Dictionary, August 2022 edition). The value in the field is a majority rule, not a fact about the parcel.

The third is that the map is not the law. Vested rights, legal nonconforming uses, variances, conditional use permits, development agreements and pending amendments all live outside the polygon layer. A building that predates the current ordinance may lawfully continue a use the map forbids, and a parcel the map permits may be unbuildable for reasons the ordinance never mentions. The National Zoning Atlas says so in its own disclaimer: the map carries no warranty of validity or accuracy, does not reflect zoning conditions in real time, may lag changes in codes, state statutes, court decisions and boundaries, depicts only what local codes say rather than state statutory overrides, is not suitable for legal, engineering or land survey purposes, and does not guarantee development feasibility, because infrastructure gaps, topography, contamination and restrictive covenants sit outside zoning entirely (National Zoning Atlas, disclaimer, read August 2026).

Where zoning data comes from: five kinds of publisher

There are five practical sources of zoning information in the United States, and they answer different questions at different levels of reliability. Knowing which one you are holding is most of the discipline.

The municipal or county GIS layer. This is the original. A planning department maintains the zoning polygons that correspond to its adopted zoning map, and in an increasing number of jurisdictions publishes them through an open data portal or a feature service. San Francisco publishes three separate layers: Zoning Districts, with 10,617 polygons carrying 120 distinct district codes and 88 distinct code section references; Height and Bulk Districts, with 5,292 polygons and 269 distinct height and bulk designations; and Special Use Districts, with 145 polygons (City and County of San Francisco Planning Department, via DataSF, all three updated August 23, 2026, queried August 24, 2026). Chicago publishes 14,929 zoning polygons carrying 1,528 distinct zone class values, with an adopting ordinance number on 7,341 of them (City of Chicago, Boundaries, Zoning Districts, current, rows updated August 10, 2026, queried August 24, 2026). This is the most authoritative source available as data, and it is still a map.

The state compilation. A handful of states aggregate local zoning into a single layer with a standardized classification. Oregon's Department of Land Conservation and Development, with support from the Oregon Department of Transportation, publishes a statewide Zoning feature service generalizing local zoning into statewide classes, built from data supplied by 229 local jurisdictions as of June 30, 2023, with non-participants marked as not shared (Oregon Department of Land Conservation and Development, statewide Zoning feature service, item created July 19, 2023). Vermont runs a formal standard: the VT Zoning GIS Data Standard, Version 2.0, dated November 12, 2025, defines a framework for digitally mapping municipal zoning districts and a process for intaking zoning map data from Regional Planning Commissions into a statewide dataset, specifying fields and treatment type domains, and explicitly not carrying setback, coverage, floor area ratio or height values (Vermont Center for Geographic Information, VT Zoning GIS Data Standard, Version 2.0, 2025). California's compilation, described above, is the largest and the most candid about its own construction.

The nonprofit atlas. The National Zoning Atlas codes zoning ordinances into standardized attributes across jurisdictions. It is the only project attempting national coverage of the regulations themselves rather than the map, and its counts are the only national numbers available.

The assessor roll. Many assessment rolls carry a zoning field, copied from the planning department at some point in the past. It is convenient, it joins to the parcel by construction, and it is the most dangerous of the five, for reasons the next sections set out.

The ordinance itself. The permitted use table, the dimensional standards, the overlay provisions and the definitions are text, usually a PDF or a hosted code library, and in most jurisdictions this is the only place the parameters exist. A lender's file that cites a zoning district without citing a code section has cited the map and called it the law.

There is a sixth category, commercial zoning products that standardize local codes into parcel-level attributes across large parts of the country. Those are a legitimate procurement option and are evaluated as a category in the companion piece on where to get parcel data. They are not a substitute for knowing what the underlying public record does and does not say, because they are built from exactly the sources listed above, and they inherit every limitation described here.

What a published zoning layer actually contains

Open the attribute table of any published zoning layer and the same pattern appears. There is an identifier, sometimes a generalized class, occasionally a date, and then nothing. The parameters are absent, not because the publishers are careless but because a polygon layer is a map of districts and the district is defined elsewhere.

Five publications, all queried through their public APIs on August 24, 2026, make the point concretely.

California's statewide compilation carries County, Jurisdiction, Code, Description, Classkey, a standardized class number and description, Source, Date, Version and a reclassification field. Oregon's carries local zone code and description, statewide zone code and description, a code reference field, an effective date, and owner name and type. Chicago carries case number, case type, zoning identifier, zone type, zone class, edit status, create and edit dates, a planned development prefix and number, comments, an ordinance number and an ordinance date. New York City's Zoning Tax Lot Database carries the borough, block and lot keys, up to four zoning districts, two commercial overlays, three special districts, a limited height district and the zoning map sheet. San Francisco carries the district code, a simplified code, the district name, a general category, a code section reference and a link, with height published as a separate layer.

Score those five against the parameters a development screen needs and the result is stark. Permitted use table: none of the five. Minimum lot size: none. Setbacks: none. Floor area ratio or lot coverage: none. Parking minimum: none. Height limit: one of five, and only because San Francisco publishes height and bulk as its own polygon layer. What the five do carry is a district identifier in every case, a generalized class in four, an effective or source date in two, an overlay or special district pointer in two, and a pointer back to the code section or adopting ordinance in two.

Two of those files are worth reading closely, because their metadata is more informative than most vendor documentation.

California's item description states that where zoning maps were not available in a GIS format, maps were converted from PDF or image maps using georeferencing techniques and then transposed onto parcel geometries sourced from county assessor data; that collection began in late 2021 and was mostly finished in late 2022 with some data updated in 2023; that contiguous areas with identical zone codes within a jurisdiction were dissolved together; and that features representing roads with right of way or null zone designations, and features smaller than four square meters, were removed. It also states that the data is provided for informational purposes only, that it should not be construed as legal advice, that users should independently verify its determinations before taking any action under the California Environmental Quality Act or any other law, and that the State of California makes no warranties as to accuracy (California Governor's Office of Land Use and Climate Innovation, California Statewide Zoning, item metadata read August 24, 2026). A statewide zoning map that is partly a georeferenced PDF transposed onto assessor geometry is still a useful screening layer. It is not a determination.

Oregon's is a study in what standardization costs. The layer holds 114,823 polygons, of which 70,574 come from cities and 44,249 from counties, across 245 distinct owner names. Every polygon carries a statewide class, drawn from 56 distinct codes. Only 653 polygons, or 0.6%, carry the local zone code, and only 649 carry the local zone description. The field named for a code reference is populated on all 114,823 records with a single value identifying the geographic names source, not a code citation. In other words, the file that makes Oregon comparable across 229 jurisdictions is the same file that severs the link from a polygon back to the ordinance section that governs it (Oregon Department of Land Conservation and Development, statewide Zoning feature service, queried August 24, 2026).

The effective dates in that same layer are worth a lender's attention. They run from May 12, 2009 to June 22, 2023. Of the 114,815 polygons carrying a date, 16,914 are dated before 2015 and a further 10,014 fall between 2015 and 2019, so 23.5% of the statewide layer carries a zoning effective date before 2020, against a compilation whose newest effective date is June 22, 2023. California's date field is worse in a different way: 38 distinct date strings appear, mixing quarter labels such as 2021 q4, calendar years as early as 2007, explicit dates in 2023, and at least one unconverted spreadsheet serial number applied to 5,844 polygons. A vintage field that has not been normalized is a warning about the whole file.

The National Zoning Atlas, and the only national counts that exist

If the map is the easy part, someone has to do the hard part. The National Zoning Atlas, run by Land Use Atlas, Inc. and begun in 2021 after the release of the Connecticut Zoning Atlas, is the only project attempting to code the regulations themselves at national scale. Its method classifies every zoning district as primarily residential, mixed with residential, or nonresidential, and extracts the types of residential uses allowed, the permitting procedures, lot development requirements such as minimum lot sizes and parking mandates, building parameters such as height caps, and permissions for accessory dwelling units, planned residential developments and mobile homes. The published method is How To Make a Zoning Atlas 2.0 (2024), and the project states that its teams code upwards of 200 regulatory characteristics for each zoning district (National Zoning Atlas, How It Works and About, read August 24, 2026).

Its counters, read on August 24, 2026, stand at more than 11,000 jurisdictions published, more than 224 million people living in published jurisdictions, and more than 1.3 million pages of zoning codes read, against a stated national scope of more than 33,000 zoning jurisdictions. Those same counters were read on August 22, 2026 during the source pack build for this pillar and stood at the same rounded values. Taken as a share, published coverage is roughly one third of the project's own stated national scope.

The statewide builds show what full coverage of a state looks like as a unit of work. The Connecticut Zoning Atlas, the first statewide atlas, analyzed all 187 zoning jurisdictions in a state of 169 municipalities and became the template; it reported that 91% of Connecticut's land was zoned for single family housing and 2% allowed multifamily (National Zoning Atlas, About, read August 24, 2026). The Vermont Zoning Atlas, built at the University of Vermont with the state's Regional Planning Commissions and coded across 2023 and 2024 under an MIT License, covers all 1,755 zoning districts in all 255 municipalities. The Texas effort is analyzing zoning conditions across all 1,478 jurisdictions, counties, cities and towns, in the state (National Zoning Atlas, Texas Zoning Atlas, read August 24, 2026).

Two cautions belong with every use of these numbers. The first is that they come from a nonprofit project, not a federal program, and should be attributed that way rather than presented as an official statistic. The second is the disclaimer, which the project publishes and which an analyst should carry rather than paraphrase away: no warranty of validity or accuracy, no real time currency, potential lag behind changes in codes, state statutes, court decisions and boundaries, coverage of local codes only rather than state statutory overrides, unsuitability for legal, engineering or land survey purposes, and an explicit statement that zoning permission does not guarantee development feasibility.

That last clause is the one lenders should read twice. It is the same boundary that separates a zoning screen from the broader question of whether a project pencils, which is treated separately in the work on the quantitative core of feasibility analysis.

There is a further wrinkle worth recording, because it shows how thin the state layer really is in places that look well served. Massachusetts runs one of the more mature state geospatial programs in the country. A title search of the ArcGIS Online catalogue for items owned by MassGIS with zoning in the title, run on August 24, 2026, returned seventeen items, all of them wellhead protection, hurricane evacuation or water supply protection zones. The single MassGIS item titled National Zoning Atlas is a web mapping application whose destination is the atlas website. A state with a strong data program can host municipal zoning layers and point at a nonprofit for the statewide picture, and that arrangement is common enough that the absence of a state zoning layer should never be read as the absence of zoning data.

An assessor use code is not a zoning district

This is the single most common error in screening work built on parcel data, and it is easy to make because the two fields sit side by side in the same table. A use code describes what the assessor believes is there now, for the purpose of levying tax. A zoning district describes what the ordinance permits, for the purpose of regulating development. They answer different questions, they are maintained by different offices, and their vocabularies are not nested.

San Francisco's assessor roll makes the divergence measurable, because it carries both fields on the same record. The Assessor Historical Secured Property Tax Rolls hold 3,934,467 rows across roll years; roll year 2025 alone holds 211,547 parcel records, with the file current as of June 26, 2026. Those 211,547 records carry 106 distinct zoning code values and 9 distinct use code values (San Francisco Office of the Assessor-Recorder, Assessor Historical Secured Property Tax Rolls, queried August 24, 2026).

Nine categories against 106. The nine are Single Family Residential with 155,111 records, Multi-Family Residential with 36,278, Miscellaneous or Mixed-Use with 5,487, Commercial Misc with 4,992, Commercial Retail with 4,377, Industrial with 2,372, Commercial Office with 1,818, Commercial Hotel with 758 and Government with 354. Each of them is spread across most of the city's zoning districts. The single use code for Single Family Residential appears in 83 distinct zoning codes, Multi-Family Residential in 77, Commercial Retail in 70, Commercial Misc in 65, Commercial Office in 63, Miscellaneous or Mixed-Use in 62, Industrial in 52, Commercial Hotel in 46 and Government in 16.

Read that the other way and the point sharpens: 1,169 parcels the assessor classifies as Single Family Residential sit in the M1 light industrial district, 561 in M2, and 497 in C2. Those are not errors in the roll. They are buildings that exist, in districts that would not permit them today, which is the ordinary condition of any city older than its current zoning ordinance. A screen that infers permitted use from an assessor use code will read those parcels as residential land. A screen that infers current use from a zoning district will read them as industrial. Both are wrong in opposite directions.

New York City reproduces the pattern at a larger scale with a different vocabulary. The Department of City Planning assigns each of its building classes to one of eleven land use categories, which is explicitly a classification of what is built rather than what is allowed. PLUTO version 26v1 holds 858,602 tax lots carrying 203 distinct primary zoning district values. The land use category for one and two family buildings covers 566,822 lots spread across 119 distinct zoning districts. Mixed residential and commercial buildings, at 56,440 lots, span 181 districts, and commercial and office buildings, at 21,252 lots, span 182 (New York City Department of City Planning, PLUTO 26v1, released May 28, 2026, queried August 24, 2026).

The practical rule is simple and worth writing into a procedure. Use codes answer questions about the existing stock and about comparables. Zoning districts answer questions about entitlement. Neither substitutes for the other, and a table that treats them as interchangeable will produce a supply estimate that cannot be defended in a credit file. The broader question of what assessor rolls do and do not record is treated in the companion work on what assessor rolls record about buildings, and the underlying variation in record quality across the country in the survey of openness and quality of parcel records by state.

The parcel is the wrong primary key

Even where a zoning value is correctly attached to a parcel, the parcel is a poor container for the answer. Districts are areas. Overlays are areas. Special purpose districts are areas. None of them are obliged to respect a lot line, and in dense cities a large minority of lots carry more than one regulatory geometry.

New York City publishes the cleanest evidence available on this, because its Zoning Tax Lot Database is built for exactly this join. Across 857,951 tax lots, updated April 5, 2026 and queried August 24, 2026, the file records 202 distinct primary zoning districts, 174 distinct secondary zoning districts and 94 distinct special districts. Of those lots, 19,666 carry a second base zoning district, 218 carry a third and 12 carry a fourth. A commercial overlay covers 74,233 lots, with a second overlay on 182 more. A special purpose district covers 108,280 lots, with a second on 321. A limited height district covers 3,045. Taken together, 185,938 lots, or 21.7% of the city, carry something beyond a single base zoning district (New York City Department of City Planning, NYC Zoning Tax Lot Database, queried August 24, 2026).

More than one lot in five, in the most heavily documented zoning environment in the country, cannot be answered by reading one district code. And where a lot is split, the primary field does not flag the ambiguity; by the city's own definition it simply records the district covering the greatest share of the lot's area.

San Francisco encodes the same problem differently. Its assessor roll resolves split zoning by inventing a combined value: 802 records in roll year 2025 carry one of 22 zoning code values that name two districts, from RC4NC3 on 235 records and RM3RM4 on 148 down to RM3/C2 on a single parcel. Those values will not join to any district table, will not match the planning department's own layer, and will silently drop out of any analysis that filters on a district list.

There is a third failure mode in the same field, and it is larger than either. The zoning code is recorded as NA on 41,942 of the 211,547 records, which is 19.8% of the roll. That gap is not random: 29,824 of those records are condominiums, with a further 1,988 live or work condominiums, 1,823 below market rate condominiums, 1,229 timeshares, 998 unsegregated cooperative units, 805 condominium economic units and 699 commercial store condominiums. A condominium record is a unit, not a lot, so the field that would carry a lot's zoning has nothing to carry. In a county where roughly one record in five is a subdivided interest rather than a parcel of land, a parcel keyed zoning join loses a fifth of the file before any analysis begins.

The remedy is not a better join. It is to treat the zoning question as a spatial intersection against the district, overlay and special district layers, to keep the parcel geometry rather than a parcel centroid where the lot is large or irregular, and to record the case where the intersection returns more than one district as a fact about the site rather than as noise. Analysts working from centroids should read the companion note on the Census geocoder against commercial services, since a point placed by street segment interpolation can land on the wrong side of a district boundary in exactly the cases that matter.

The seven parameters a development screen actually needs

A zoning code can run to hundreds of pages, and a screening exercise does not need most of it. What a lender or a developer needs before spending money on design is a small, stable set of parameters, each of which either permits the contemplated project or kills it. Everything else is refinement.

The seven are these.

The permitted use table. Is the contemplated use allowed as of right, allowed conditionally, or not allowed? This is the only binary in the set, and it is the one that ends screens. Conditional approval is not permission; it is a process with a calendar and a discretionary outcome, and a screen should record it as such.

Minimum lot size. Both the floor for a buildable lot and, in many residential districts, the effective density control. Where a district sets a minimum lot area per dwelling unit, this parameter and not the floor area ratio determines unit count.

Setbacks and yards. Front, side and rear. These carve the buildable footprint out of the lot and interact with lot depth and frontage in ways that punish irregular parcels. San Francisco sets its basic rear yard at 30% of total lot depth, never less than 15 feet, in the RH, RM-1, RM-2, RTO-1 and RTO-M districts, and 25% elsewhere in the listed districts, and it caps the required front setback at 10 feet, or 15 feet on block faces where more than 75% of properties and both adjacent parcels are set back 15 feet or more (San Francisco Planning Code Sections 132 and 134, with Section 132 last amended by Ordinance 10-26 effective March 16, 2026).

Height. Usually a number of feet, sometimes a number of stories, and always with a measurement convention attached. San Francisco fixes the measurement point at curb level on the fronting street where the lot is level with or slopes down from the street, applied over a lot depth of up to 100 feet or to the midblock line, whichever is greater (San Francisco Planning Code Section 260(a), 2026). A height limit without its measurement rule is not a parameter, it is a number.

Floor area ratio or lot coverage. The bulk control. Some codes use one, some use both, and some disapply them for the most common uses, which is a trap examined below.

Parking minimums where they still exist. A parking requirement converts directly into land and cost, and the last decade of reform means the answer is jurisdiction specific and recently changed in many places. Screening on a stale parking minimum overstates land need; assuming reform overstates capacity.

Overlay and special districts. The layer that rewrites the base rules, and the one most often missed because it lives in a separate map and a separate article of the code. In New York City alone, overlays and special districts touch 21.7% of tax lots.

Two things follow. The first is that these seven parameters, plus the district identifier, are what a screening database should hold, and every one of them needs a code citation with a section number and an effective date attached, not a free text note. The second is that six of the seven are absent from every published zoning polygon layer examined for this article. The map gets you the district. The parameters require a document.

Once the seven are in hand, they become inputs rather than answers. Turning them into a building envelope, a unit count or a rentable area is a separate exercise with its own failure modes, set out in the companion piece on estimating buildable area from a parcel, and constrained further by site conditions such as slope, which is treated in the work on slope and cut-fill screening.

Reading the code: the verification chain

The discipline that makes zoning data usable in a credit file is a chain, and each link can be checked by someone else. Map, then text, then the planning office. Skipping a link is what produces a screen that reads well and does not survive diligence.

Link one, the map. Intersect the site geometry with the district layer, the overlay layer and the special district layer, from the jurisdiction's own publication where one exists. Record the district code, every overlay and special district returned, the layer's publication date, and whether the intersection returned more than one district. Record the source as the jurisdiction, not as an aggregator.

Link two, the text. Take the district code to the ordinance and pull the seven parameters, citing section numbers. Read the definitions article first, because gross floor area, lot, height and setback are defined terms and the definitions differ between codes. Read the exceptions, because they are where the parameters go to die.

Link three, the planning office. Confirm the district, confirm that no amendment is pending on the parcel, and confirm the status of any nonconforming use or prior approval. This is a phone call or a zoning verification letter, and it is the only link that returns a statement the jurisdiction will stand behind.

San Francisco supplies a clean worked illustration of why link two cannot be skipped, because its code, its assessor roll and its zoning map are all public. Start with the roll year 2025 file: 211,547 parcel records. Of those, 169,605 carry a zoning code at all, the remaining 41,942 being the records described above, most of them condominium and cooperative interests rather than lots. Of the 169,605, some 142,645 sit in one of the districts for which Planning Code Table 124 sets a basic floor area ratio, a table running from 1.0 to 1 in the RED and RED-MX districts to 9.0 to 1 in C-3-O, last amended by Ordinance 245-25 effective January 12, 2026.

At that point a screen built on the map alone would report that 67.4% of the city's parcel records have a published floor area ratio limit. Then you read the next subsection. Section 124(b) exempts residential uses in R, RC, NC and Mixed Use Districts from floor area ratio limits altogether. Apply that single subsection and the count of records whose tabulated floor area ratio actually binds falls from 142,645 to 8,941, which is 4.2% of the roll (San Francisco Office of the Assessor-Recorder, roll year 2025, queried August 24, 2026; San Francisco Planning Code Table 124 and Section 124(b), 2026).

That is not a San Francisco quirk. It is the general shape of the problem. The map gave a district to 80.2% of records. The table gave a number to 67.4%. One subsection of the same code took the number back from almost all of them, and for those parcels the binding controls are height, rear yard and dwelling unit density instead, each of which sits in a different article of the same document. No polygon layer, no matter how current, contains that sequence. It is a reading exercise, and its output is a citation.

The same logic applies to the exceptions that expand rather than restrict. Oregon's statewide layer metadata flags that House Bill 2001 of 2019 requires certain cities to allow middle housing in residential zones, so the mapped residential classes may understate what is permitted (Oregon Department of Land Conservation and Development, statewide Zoning layer metadata, 2023). A state statutory override does not appear on a local map, and the National Zoning Atlas says the same of its own coverage. When a state preempts local zoning, every layer built from local codes is out of date until each jurisdiction amends, and the parcels affected are exactly the ones a housing screen cares about.

Coverage, honestly

Every coverage claim about zoning data is a claim about a denominator, and the denominators available are not comparable. Six of them are in circulation and each is defensible on its own terms.

Count general purpose local governments and you get 38,736 (U.S. Census Bureau, 2022 Census of Governments, released 2023). Count zoning jurisdictions and you get the National Zoning Atlas figure of more than 33,000, which no federal source corroborates. Count jurisdictions whose codes have been read and coded and you get the atlas's own figure of more than 11,000, roughly a third of its stated scope. Count states with a published statewide zoning layer and the number is small enough to name individually. Count parcels covered by some zoning polygon and you get a very large number that says nothing about whether the parameters exist anywhere in structured form. Count population and you get the atlas's 224 million, which is the most flattering framing available because zoning coverage correlates with density.

The honest formulation is a two part statement. Zoning polygon coverage is wide and improving: most metropolitan jurisdictions publish a district layer, several states now aggregate them, and a national compilation of maps is a tractable data engineering problem. Zoning parameter coverage is narrow, unstandardized and concentrated in one nonprofit project's output plus whatever a given jurisdiction happens to publish. The two facts are usually reported as one number, and the number reported is always the first one.

Three practical consequences follow for anyone building a screen at scale.

First, the vintage of a zoning polygon is a property of the polygon, not of the file. Oregon's layer, compiled in 2023, carries zoning effective dates back to 2009 on 14.7% of its polygons. California's carries free text dates spanning 2007 to 2023 in 38 different formats. A layer downloaded today can contain a district boundary adopted sixteen years ago, and the file's own update date will not tell you.

Second, standardized classes and local codes are different products and should be stored as different fields. Oregon's 56 statewide classes and California's 14 standardized classes are what make cross jurisdiction comparison possible. The 12,091 local codes in California's northern file and 8,886 in its southern file are what make a code lookup possible. A pipeline that keeps only the standardized class has bought comparability and sold traceability.

Third, the tail is not small. Chicago's current layer carries 1,528 distinct zone class values, of which 1,446 belong to the Planned Development zone type, covering 1,459 of 14,929 polygons. Nearly one in ten zoning polygons in Chicago is governed by an individually adopted ordinance rather than by a district rule, and 7,341 of the city's polygons carry an ordinance number precisely because the rules are in the ordinance and not in the district (City of Chicago, Boundaries, Zoning Districts, current, queried August 24, 2026). Planned developments, planned unit developments, form based overlays and development agreements are the growth area of American land use regulation, and they are structurally resistant to standardization, because each one is bespoke by design.

A reproducible screening sequence

What follows is the sequence this article's evidence supports, written so that a second analyst can repeat it and reach the same answer. It assumes a site, a contemplated use and a lending or investment decision that has not yet spent money on design.

Step 1. Identify the jurisdiction, not the county. Determine whether the site is inside a municipality, in unincorporated county territory, or in a township with independent zoning authority. Getting this wrong sends the entire screen to the wrong ordinance. Failure mode: a mailing address inside a city's postal area for a site in unincorporated county land.

Step 2. Pull the jurisdiction's own zoning layer. Prefer the planning department's published service over any aggregation. Record the layer name, the publisher, the update date and the query date. Failure mode: an aggregated statewide class that cannot be taken back to a code section, as in Oregon's layer where 99.4% of polygons carry no local code.

Step 3. Intersect geometry, not a point. Use the parcel polygon. Capture every district, overlay, special district and height district that the site touches, and flag multi district results explicitly. Failure mode: a centroid that misses an overlay covering the frontage.

Step 4. Cross check against the assessor roll, and expect disagreement. A roll zoning value that differs from the planning layer is information, not an error to be reconciled away. San Francisco's roll carries 106 zoning values while the planning department's layer carries 120. Failure mode: trusting the roll because it joined cleanly.

Step 5. Read the code for the seven parameters. Definitions first, then the use table, then the dimensional table, then the exceptions. Cite section numbers and effective dates. Failure mode: recording a floor area ratio from a table without reading the subsection that disapplies it.

Step 6. Check for overrides. State preemption statutes, recent local amendments and pending applications. Failure mode: screening a residential site against a local map in a state that has preempted it.

Step 7. Verify with the planning office before the number leaves the building. A zoning verification letter or a documented conversation, dated and attributed. Failure mode: a screen that was never tested against the only party whose opinion binds.

Step 8. Record provenance for every parameter. Source, section, effective date, read date and reader. This is what allows the screen to be re-run in twelve months and compared. The general case for this discipline is made in the piece on why every number needs a source and a date, and the ordering of this screen against the rest of a public records workflow in the note on public-records checks before paying for reports and the 30-minute pre-term-sheet site screen.

Where this method stops

A zoning screen answers one question: what the ordinance, as published, appears to permit on this ground today. It does not answer whether a project is feasible, financeable or approvable, and the honest boundaries are worth stating.

It does not establish legal permission. Only the jurisdiction can do that, and only in writing. Nothing in a polygon layer or a coded atlas is a determination, and every publisher examined for this article says so in its own terms.

It does not capture private restrictions. Restrictive covenants, easements, reciprocal easement agreements and deed restrictions bind land independently of zoning and appear in the title record rather than the ordinance.

It does not capture physical constraints. Floodplain, wetlands, slope, contamination, utility capacity and access are separate screens with separate sources, and a parcel can be perfectly zoned and undevelopable. The atlas disclaimer names exactly this gap.

It does not capture process risk. A conditional use, a variance, a rezoning or a planned development approval is a calendar and a political question, and the data will not price it.

It does not capture vested rights and nonconformities, which is why the 1,169 San Francisco parcels the assessor calls Single Family Residential inside the M1 district are neither an error nor an opportunity until someone reads the file.

MMCG Analytics is a map first commercial real estate analytics platform built by MMCG Invest, LLC in San Francisco for lenders and investors, and it is built on federal, state and public data sources with source and vintage provenance carried on displayed values. Its analytical layers include parcels, flood, wetlands, wind risk, terrain, traffic and demographics, sourced from public records. Zoning is not among them, and this article explains why that is a defensible position rather than a gap: a national zoning layer that carried the parameters would have to be a national reading project, not a national mapping project, and the only organization attempting it publishes its own coverage counters and its own disclaimer. MMCG provides data and analytics; the lending decision, and the zoning determination behind it, rest with the lender and the jurisdiction.

Frequently asked questions

Is there a national zoning database for the United States?

No federal agency publishes one. Zoning is a power states delegate to local governments, there is no federal mandate and no national standard schema in force, and the statewide layers that exist are compilations of local maps assembled by a state agency or a university. The nearest thing to a national resource is the National Zoning Atlas, a nonprofit project that codes zoning ordinances into standardized attributes; read on August 24, 2026 its counters stood at more than 11,000 jurisdictions published against a stated national scope of more than 33,000 zoning jurisdictions, with more than 1.3 million pages of zoning codes read. Those are the only published national counts this research found in a public source, and they are a nonprofit's figures rather than a federal statistic.

How do I find the zoning for a specific parcel?

Identify the jurisdiction that actually zones the site, which may be a municipality, an unincorporated county area or a township, then pull that jurisdiction's own zoning layer rather than an aggregation. Intersect the parcel polygon, not a centroid, with the district layer and with any overlay, special district and height district layers, and record every result. Then take the district code to the ordinance and read the seven parameters that matter: permitted uses, minimum lot size, setbacks, height, floor area ratio or coverage, parking, and overlays. Finish with the planning office, because only the jurisdiction can confirm the district, the absence of a pending amendment and the status of any nonconforming use.

What is the difference between a zoning district and an assessor use code?

A use code records what the assessor believes exists on the parcel today, for tax purposes. A zoning district records what the ordinance permits, for regulatory purposes. They are maintained by different offices and their vocabularies are not nested. San Francisco's roll year 2025 file carries 9 distinct use codes and 106 distinct zoning codes on the same 211,547 parcel records, and the single use code for Single Family Residential appears in 83 different zoning codes, including 1,169 records in the M1 light industrial district. Use codes answer questions about the existing stock and about comparables; zoning districts answer questions about entitlement.

Does zoning GIS data include setbacks, height limits and floor area ratio?

Almost never. Of five published zoning layers examined on August 24, 2026, covering California statewide, Oregon statewide, Chicago, New York City and San Francisco, none carried a permitted use table, a minimum lot size, a setback, a floor area ratio or a parking requirement as an attribute. One of the five carried a height limit, and only because San Francisco publishes height and bulk as a separate polygon layer. What the layers do carry is a district identifier in every case, a generalized class in four, an effective or source date in two, and a pointer back to a code section or adopting ordinance in two. The parameters live in the ordinance text.

How current is zoning GIS data?

Currency is a property of the individual polygon, not of the file. Oregon's statewide layer, compiled in 2023, carries zoning effective dates from May 12, 2009 to June 22, 2023, with 16,914 of its 114,815 dated polygons effective before 2015 and a further 10,014 between 2015 and 2019. California's statewide compilation records dates in 38 distinct formats, mixing quarter labels, bare years as early as 2007, explicit 2023 dates and at least one unconverted spreadsheet serial number applied to 5,844 polygons. A file update date says when the compilation was refreshed, not when the zoning on a given parcel was adopted.

What is split-lot zoning and how should a screen handle it?

It is a parcel that falls in more than one zoning district, or that carries an overlay or special district on part of its area. It is common. Across 857,951 New York City tax lots, 19,666 carry a second base zoning district, 74,233 carry a commercial overlay, 108,280 sit in a special purpose district and 3,045 in a limited height district, so 185,938 lots, 21.7% of the city, cannot be answered by one district code. New York's own field definition records the district covering the greatest share of the lot's area when a lot is split, which means the primary value is a majority rule rather than a fact. Handle it by capturing every district returned by the intersection and flagging the multi district case explicitly.

Can a zoning map support a lending decision on its own?

No. A polygon layer places a site in a district; it does not establish what the district permits, and every publisher examined says so. San Francisco illustrates the gap: 142,645 of 211,547 parcel records on the roll year 2025 file sit in a district for which Planning Code Table 124 sets a basic floor area ratio, but Section 124(b) exempts residential uses in the R, RC, NC and Mixed Use Districts, so the tabulated ratio actually binds on 8,941 records, or 4.2%. The map is evidence; the ordinance is the rule; the planning office is the only party whose statement binds. A screen that stops at the map has cited a map and called it the law.

Sources

1. U.S. Census Bureau, 2022 Census of Governments, Organization component, Table 2, Local Governments by Type and State, released 2023; counts of county, municipal, township, special district and independent school district governments. https://www.census.gov/data/tables/2022/econ/gus/2022-governments.html
2. National Zoning Atlas (Land Use Atlas, Inc.), home page coverage counters and stated national scope of more than 33,000 zoning jurisdictions, read August 24, 2026. https://www.zoningatlas.org/
3. National Zoning Atlas, About the National Zoning Atlas: project start in 2021, upwards of 200 regulatory characteristics coded per zoning district, the 2021 Connecticut Zoning Atlas and its 91% and 2% land share findings, read August 24, 2026. https://www.zoningatlas.org/about
4. National Zoning Atlas, How the National Zoning Atlas is Made, including the district classification scheme and the published method How To Make a Zoning Atlas 2.0 (2024), read August 2026. https://www.zoningatlas.org/how
5. National Zoning Atlas, Disclaimer: no warranty of validity or accuracy, no real time currency, local codes only rather than state statutory overrides, unsuitability for legal, engineering or land survey purposes, and the statement that zoning permission does not guarantee development feasibility, read August 2026. https://www.zoningatlas.org/disclaimer
6. National Zoning Atlas, Texas Zoning Atlas: analysis across all 1,478 jurisdictions in Texas, read August 24, 2026. https://www.zoningatlas.org/texas
7. Xu, Markley, Bronin and Drogaris, A National Zoning Atlas to Inform Housing Research, Policy, and Practice, Cityscape Vol. 25 No. 3, symposium 100 Years of Federal-Model Zoning, U.S. Department of Housing and Urban Development, Office of Policy Development and Research, November 2023. https://www.huduser.gov/portal/periodicals/cityscape/vol25num3/article3.html
8. Vermont Center for Geographic Information, VT Zoning GIS Data Standard, Version 2.0, dated November 12, 2025: the statewide framework for digitally mapping municipal zoning districts, its field and domain specification, and its exclusion of setback, coverage, floor area ratio and height values. https://files.vcgi.vermont.gov/other/standards-guidelines/zoning/zoning-standard.html
9. University of Vermont with the Vermont Regional Planning Commissions, Vermont Zoning Atlas: all 1,755 zoning districts across all 255 municipalities, coded 2023 to 2024, MIT License. https://verso-uvm.github.io/Vermont-Zoning-Atlas/
10. Oregon Department of Land Conservation and Development with Oregon Department of Transportation support, statewide Zoning feature service, ArcGIS Online item metadata, owner OR_FRAMEWORK, item created July 19, 2023, data from 229 local jurisdictions as of June 30, 2023, read August 24, 2026. https://www.arcgis.com/sharing/rest/content/items/72416393747c4cdf9110920656f90f83?f=json
11. Oregon Department of Land Conservation and Development, statewide Zoning feature service, layer ZoneOR_Gov2Pub: field list, polygon count, statewide class distribution, local code population and effective date distribution, queried August 24, 2026. https://services8.arcgis.com/8PAo5HGmvRMlF2eU/arcgis/rest/services/Zoning/FeatureServer/0
12. California Governor's Office of Land Use and Climate Innovation, California Statewide Zoning North, ArcGIS feature service and item metadata: collection from 535 of California's 539 jurisdictions, georeferencing of PDF and image maps onto assessor parcel geometry, collection 2021 to 2023, field list, polygon count, local code and standardized class counts and date field distribution, published 2026, queried August 24, 2026. https://services8.arcgis.com/BcP3FB8Ua7DeCr93/arcgis/rest/services/California_Statewide_Zoning_North__19162237743447578111/FeatureServer/0
13. California Governor's Office of Land Use and Climate Innovation, California Statewide Zoning South, ArcGIS feature service: polygon count, jurisdiction and county counts and local code count, published 2026, queried August 24, 2026. https://services8.arcgis.com/BcP3FB8Ua7DeCr93/arcgis/rest/services/California_Statewide_Zoning_South_19904491431473492211/FeatureServer/0
14. City and County of San Francisco Planning Department, Zoning Map, Zoning Districts, DataSF dataset 3i4a-hu95, updated August 23, 2026, queried August 24, 2026. https://data.sfgov.org/resource/3i4a-hu95.json
15. City and County of San Francisco Planning Department, Zoning Map, Height and Bulk Districts, DataSF dataset h9wh-cg3m, updated August 23, 2026, queried August 24, 2026. https://data.sfgov.org/resource/h9wh-cg3m.json
16. City and County of San Francisco Planning Department, Zoning Map, Special Use Districts, DataSF dataset 5yf5-ms5f, updated August 23, 2026, queried August 24, 2026. https://data.sfgov.org/resource/5yf5-ms5f.json
17. San Francisco Office of the Assessor-Recorder, Assessor Historical Secured Property Tax Rolls, DataSF dataset wv5m-vpq2, 3,934,467 rows, roll year 2025 current as of June 26, 2026, queried August 24, 2026. https://data.sfgov.org/resource/wv5m-vpq2.json
18. San Francisco Planning Code, Articles 1, 1.2 and 2.5, hosted by American Legal Publishing: Section 102 definitions, Table 124 basic floor area ratio limits and Section 124(b) residential exemption (Table 124 last amended by Ordinance 245-25, effective January 12, 2026), Sections 132 and 134 front setback, rear yard and lot coverage (Section 132 last amended by Ordinance 10-26, effective March 16, 2026), and Sections 250 and 260 height and measurement, 2026. https://codelibrary.amlegal.com/codes/san_francisco/latest/sf_planning/0-0-0-49051
19. City of Chicago, Boundaries, Zoning Districts (current), Chicago Data Portal dataset dj47-wfun, rows updated August 10, 2026: polygon count, zone class and zone type distribution, planned development counts and ordinance field population, queried August 24, 2026. https://data.cityofchicago.org/resource/dj47-wfun.json
20. New York City Department of City Planning, NYC Zoning Tax Lot Database, NYC Open Data dataset fdkv-4t4z, updated April 5, 2026: tax lot count, zoning district, commercial overlay, special district and limited height district field population, queried August 24, 2026. https://data.cityofnewyork.us/resource/fdkv-4t4z.json
21. New York City Department of City Planning, Primary Land Use Tax Lot Output (PLUTO), version 26v1, released May 28, 2026: lot count, primary zoning district counts, land use category distribution and the published field definitions for primary zoning district, split zone and land use category, queried August 24, 2026. https://data.cityofnewyork.us/resource/64uk-42ks.json
22. New York City Department of City Planning, PLUTO Data Dictionary, August 2022 edition (22v2), including Appendix D land use categories. https://www.nyc.gov/assets/planning/download/pdf/data-maps/open-data/PLUTODD.pdf
23. Federal Geographic Data Committee, Geographic Information Framework Data Content Standard, Part 1: Cadastral, FGDC-STD-014.1-2008, May 2008: the federal content standard for parcel data, which does not address land use. https://www.fgdc.gov/standards/projects/framework-data-standard/GI_FrameworkDataStandard_Part1_Cadastral.pdf
24. ArcGIS Online content catalogue, owner scoped title search for MassGIS items containing zoning, run August 24, 2026: seventeen items returned, none a zoning district layer, and one MassGIS item titled National Zoning Atlas published as a web mapping application. https://www.arcgis.com/sharing/rest/search
25. MMCG Research, U.S. Zoning Data Series: the seven parameter scoring of five published zoning layers, the San Francisco roll year 2025 verification chain counts and split district value tally, the distinct code, polygon, vintage bucket and use code span counts computed from sources 11 to 21, and all derived shares. Computed August 24, 2026; MMCG database, 2026. https://mmcganalytics.com/methodology/