How we counted NYC's cars.
Plain language on purpose. The point of publishing a method is that somebody can check it, and a method nobody can follow is a claim rather than a method.
Every figure in this section comes from two government sources: New York State DMV's vehicle registration file for the five New York City counties, and NHTSA's federal VIN decoder, which supplies the model name DMV does not publish. Nothing is modelled, estimated, sampled or bought from a data vendor. The August 2026 snapshot downloaded 2,208,957 records, of which 2,124,852 were currently registered vehicles.
- Records downloaded
- 2,208,957
- In scope
- 2,181,048
- Currently registered
- 2,124,852
- Snapshot id
- 20260817T091745Z
Data: August 2026 snapshot
Where the data comes from.
The source is a single published file: Vehicle, Snowmobile, and Boat Registrations (NYS DMV), dataset w4pv-hbkt on New York State Open Data (data.ny.gov), last republished by the state on 2026-08-02 and downloaded on 2026-08-17.
It is a statewide file of registration records, republished monthly. We take the five counties that make up New York City, filtered at the source, so only New York City records are ever transferred. The file is public and anybody can download the same one.
| DMV county | Borough | Registrations |
|---|---|---|
| Kings | Brooklyn | 526,159 |
| Queens | Queens | 814,524 |
| Bronx | Bronx | 265,112 |
| New York | Manhattan | 231,183 |
| Richmond | Staten Island | 287,874 |
Why a VIN decoder is needed at all.
DMV publishes the make as a five-letter code and the model year, and it does not publish the model. There is no column anywhere in the file that says Camry or RAV4, so the model name has to be read out of the VIN.
The first eight characters of a VIN describe the manufacturer and the vehicle it built: make, model line, body style, engine and restraint system. The tenth character is the model year. Those nine characters are exactly what a decoder needs, and they are the only nine characters this analysis ever handles. The decoder is NHTSA's vPIC, a federal service, and its answers are public domain.
What we do with VINs, and what we refuse to do.
We never download a full VIN. The request sent to the state portal asks for characters 1 to 8 and character 10, and nothing else.
- Characters 11 to 17 are the plant code and the serial number, the part that makes a VIN unique to one car. They are never transferred.
- No full VIN is stored, exported, logged, or committed to source control. The database schema forbids storing more than eleven characters.
- The nine characters that are kept describe a factory configuration shared by tens of thousands of vehicles. They cannot identify one.
- We do not request the registrant's city, the licence plate, or any suspension or revocation flag. This is a project about vehicles, not owners.
What is counted, and what is filtered out.
Boats, trailers and snowmobiles are removed, which took 27,909 records out of the 2,208,957 downloaded. Of the 2,181,048 vehicle records that remain, 2,124,852 are currently registered and are what every headline figure in this section counts.
| Record type | Records |
|---|---|
| Boat registration | 14,748 |
| Trailer registration | 13,149 |
| Snowmobile registration | 12 |
| Status | Records | Share |
|---|---|---|
| Currently registered | 2,124,852 | 97.4% |
| Expired but still in the file | 16,057 | 0.7% |
| No usable expiration date | 40,139 | 1.8% |
A record with no usable expiration date is called unknown. It is not assumed to be active and it is not counted in any headline figure. Motorcycles, buses and heavy trucks are kept: they are road vehicles and they are genuinely registered here, so they appear in the make rankings. Where a table is limited to cars and light trucks, it says so and names the body types it used. The classics study is the one that does this most, because most of what it would otherwise count is motorcycles.
How the model name is worked out.
A model name was identified for 2,102,828 of the 2,124,852 currently registered vehicles, leaving 22,024 counted at the make level with the model left blank.
Every registration is reduced to its nine-character decode key, and identical keys are collapsed, because most vehicles in New York are one of a few thousand factory configurations. Each distinct key is sent to the federal decoder once and the answer is cached, so the whole city is decoded in under two thousand requests rather than two million. Where the decoder returns a make but no model, the registration is counted at the make level and marked as having no model. Where it returns nothing, DMV's own make code identifies the manufacturer if that code is unambiguous, and the model is left blank. Nothing is filled in by inference.
The seventeen-character VIN became a federal standard for model year 1981. Before that, every manufacturer used its own scheme, so there is no grammar for a decoder to read. Across model years 1960 to 1989, a model was read for 6,840 of the 7,412 registrations from 1981 on, and for 163 of the 9,287 before it. Those few are the modern rule applied to numbers that predate it, so they are counted and never published as models.
Which model year we use.
Where DMV's model year and the year encoded in the VIN disagree, DMV's is used.
It is the year printed on the registration document and the state stands behind it. The VIN's tenth character repeats on a thirty-year cycle, so on its own it cannot tell a 1996 vehicle from a 2026 one. The disagreement rate is recorded on every run.
Taxis are counted separately.
The 104,465 active Taxi and Limousine Commission licences are never added to the registration counts on these pages.
A vehicle licensed by the TLC is also registered with DMV, so adding the two together would count it twice. More to the point, a for-hire vehicle and a family car are different things, and blending them would make both numbers mean less.
What these numbers do not tell you.
Stated plainly, because they change how the figures should be read.
A registration is not a car on the road
It is a record in a state database. A registered vehicle may be off the road for a year, and a car driven every day in Brooklyn may be registered at an address in New Jersey and appear nowhere in this count.
The borough is a mailing address, not a parking spot
It is the county on the registration, which is where the registrant receives mail. A company that registers a fleet at one headquarters puts all of it in one borough regardless of where those vehicles work.
The file is a rolling two-year window
DMV drops registrations that expired more than two years ago, so this is not a census of every vehicle ever registered here. Long-expired vehicles are not in the file at all.
The decoder does not know everything
Coverage is thinner for older vehicles and for low-volume manufacturers. Where a model cannot be read, the registration is reported as undecoded rather than dropped or guessed at, which is what keeps the percentages honest.
Clerks and dealers type these records
Make codes, body types and model years are entered by hand. An aggregate over two million records is robust to that; a single row is not, and nothing here should be read as a fact about one specific vehicle.
Model families, not trims
“Toyota Camry” covers every Camry. Trim detail is not reliably recoverable from the first eight VIN characters for every manufacturer, so it is not claimed.
Every figure carries a snapshot date
DMV republishes monthly. Figures from different snapshots should not be compared without saying so, which is why the August 2026 label appears on every page in this section.
Snapshots, and how often this refreshes.
Every figure on these pages comes from one dated snapshot, 20260817T091745Z, and prior snapshots are never overwritten.
Each downloaded file is recorded with its size, its row count and a checksum, so a published figure can be traced back to the exact file it came from. The state republishes the registration file monthly and the intent is to refresh these pages against it, republishing the whole set together rather than updating one table at a time. When that happens the snapshot label at the top of each page changes with the numbers, and a figure quoted from an older snapshot stays traceable to the one it came from.
Attribution.
Vehicle registration data: New York State Department of Motor Vehicles, via the New York State Open Data portal. VIN decoding: National Highway Traffic Safety Administration, Product Information Catalog and Vehicle Listing (vPIC), a United States government work in the public domain. For-hire vehicle data: New York City Taxi and Limousine Commission, via NYC Open Data.
This analysis is not produced by, endorsed by, or affiliated with any of those agencies.
New York State DMV vehicle registration file, August 2026 snapshot of the five New York City counties. Model names come from NHTSA's federal VIN decoder. A registration is a record in a state database, not a car on the road.
Cite as: Nacmias Auto analysis of NYS DMV registration data, Aug 2026. Reuse with attribution is welcome.