Where the data comes from

The data source

FARS records every crash on a public road in the 50 states, the District of Columbia and Puerto Rico in which someone died. Analysts in each state code the details from police crash reports, death certificates, vehicle registration files, hospital and emergency medical records, and other state records. This site covers the 50 states and the District of Columbia.

The raw files are published by NHTSA and are US government works, free to reuse. FARS data explained describes what is in them, how to get them and the tables you can download from this site.

Years covered

Every measure starts in the first year the federal data can support it. For the most recent years (2020–2024) the site loads every crash, vehicle and person record; earlier years are loaded as national and state totals only.

Data last updated 30 September 2026.

What counts as a fatal crash

A crash is included when it involved a motor vehicle travelling on a road normally open to the public, and at least one person died within 30 days of it. Someone who dies on day 31 or later is not counted. The person who died can be anyone involved: a driver, a passenger, a pedestrian, a cyclist or someone else outside a vehicle.

Crashes on private property, such as driveways and parking lots, and deaths caused by a medical event rather than by the crash itself, are not included. Crashes in which nobody died are not in FARS at all, so this site cannot say how many crashes happen overall.

Definitions

Deaths

People who died within 30 days as a result of the crash.

Fatal crashes

Crashes with at least one death. One crash can cause several deaths, so deaths are usually higher than fatal crashes.

Pedestrians

People on foot who were killed. People using wheelchairs, skateboards or scooters are recorded separately and are not included.

Cyclists

People riding bicycles or other pedal cycles who were killed in a crash with a motor vehicle.

Motorcyclists

Riders and passengers of motorcycles, mopeds and other motorized cycles who were killed.

Alcohol-impaired driving deaths (BAC .08+)

Everyone killed in a crash where at least one driver had a blood alcohol concentration of .08 g/dL or higher. Many drivers are never tested, so NHTSA estimates the missing results ten times over and averages the counts; this site uses the same estimates, so its totals match NHTSA's published figures. Used for the drunk driving pages and the "Deaths by type" tables.

Police-reported drinking driver

Police reported that at least one driver had been drinking. This comes from the police report, not from test results, so it counts fewer crashes than the estimate above. It is used where crashes are filtered or split by time and place (the "alcohol" filter, pattern charts and "Crash factors"), because the estimates cannot be split that way. It does not mean alcohol caused the crash.

Speeding-related

At least one driver was recorded by police as racing, going too fast for conditions or exceeding the speed limit. Speeding-related deaths are everyone killed in those crashes. It does not mean speeding caused the crash, or that the person who died was speeding.

Distraction-affected

Police identified at least one driver as distracted at the time of the crash, including drivers recorded as inattentive or careless, or, from 2018, recorded a distracted driver of a vehicle that was not in the collision. This is NHTSA's definition, and the totals match its published figures from 2012, when the current coding began. Distraction is hard to establish after a crash and states record it differently, so these counts are likely too low and vary between states for reasons other than driving. It does not mean distraction caused the crash, or that the person who died was the distracted driver.

At night

The crash happened between 6pm and 6am (18:00 to 05:59).

At weekends

The crash happened on a Saturday or Sunday.

Deaths inside a vehicle and other road users

On vehicle pages, deaths inside are people riding in that vehicle. Other road users are people killed in the same crashes elsewhere: in other vehicles, on foot or on a bicycle.

Where a detail was not reported or is unknown, that crash is left out of the chart or share that needs the detail, but it is still counted in the totals.

Rates and population

Death rates are deaths per 100,000 residents, using US Census Bureau population estimates for each state and county in the same year. Estimates are available from 2000, so rates start then.

City rates use the Census Bureau's city and town estimates, from 2020. FARS records cities with its own codes rather than Census ones, so each city is matched to the Census place of the same name in the same county, or, where FARS means a town rather than a village of the same name (as in parts of New York), to the town. Police sometimes record a crash under the nearest city even when it happened just outside the city limits, so a city's rate can be higher than the deaths of its own residents would suggest, especially for small cities next to busy roads. Where a city's share of its county's deaths is more than 4 times its share of the county's population, FARS is evidently recording a wider area under the city's name, so the city page gives deaths but no rate. Places without an incorporated Census population, such as many unincorporated communities, have no rate.

Small places have few deaths, so a single year's rate can jump around. Where a county's rate rests on fewer than 20 deaths the page says so, and county and state pages also give an average rate over several years: all the deaths in those years divided by all the residents, rather than an average of the yearly rates.

Per-person rates say nothing about how much people drive. A rural county where residents drive long distances can have a high rate even if its roads are no more dangerous per mile. For the nation and each state, pages also give deaths per 100 million vehicle miles traveled, the rate NHTSA publishes, using the Federal Highway Administration's estimates of miles driven (Highway Statistics tables VM-1 and VM-2): nationally from 1975 and by state from 1980. FHWA publishes a year's miles several months after the FARS file, so the newest year can appear without a per-mile rate for a while. Miles driven are not estimated for counties or cities.

Rankings count the places with strictly more deaths, or a strictly higher rate, so places that tie share a rank.

How new data arrives

NHTSA publishes each year's FARS file more than a year after the year ends, once states have finished their records. It publishes a final version of that year about a year later, alongside the next year, which can change the counts slightly. The latest year shown here is usually one to two years behind the current date.

This site checks NHTSA every week for a new year or a final file. New data is imported and checked against the year before, or the version it replaces: totals within a normal range, every state present, blood alcohol estimates and Census population loaded. It goes live only when every check passes; otherwise it is held back for review. When a final file replaces an earlier one, crashes it no longer contains are removed.

Older years

Years before the detailed data are loaded as national and state totals, straight from NHTSA's files for each year. The codes FARS uses for people and vehicles changed over the years (pedestrians and cyclists were recoded in 1982, motorcycles in 1982 and 1991, speeding is recorded for each vehicle only from 2009, and distraction is comparable only from 2012), so each period is counted with its own codes. The results were checked against NHTSA's published national totals.

Location accuracy

Crash locations are the latitude and longitude recorded in the police report. They can be approximate, and some crashes have no coordinates at all. Those crashes are still counted in every total but do not appear on maps.

Each crash is assigned to a state, a county and, where it happened inside city limits, a city, using FARS geographic codes. Crashes outside city limits appear on county and state pages only. Cities get a page of their own once they have at least 10 fatal crashes across the loaded years. Road names are shown as written in police reports, so the same road can appear under different names.

County and state outlines on the maps are simplified US Census Bureau boundaries and are for orientation only.

Vehicle data

Makes and models come from NHTSA's decoding of each vehicle's identification number (VIN) using its vPIC service. Vehicles whose VIN could not be decoded are counted in every crash total but are not included on make and model pages.

How often a vehicle appears in fatal crashes mostly reflects how many of them are on the road and how they are used. Popular pickups and sedans appear often because there are so many of them. These figures are not a safety rating. For crash test results, see the ratings published by NHTSA and the Insurance Institute for Highway Safety. Models and model years get a page of their own once they appear in at least 10 fatal crashes.

Caveats

• Small places have few crashes, so their numbers can swing sharply from year to year. A change from 2 deaths to 4 is a 100% rise but may not mean anything has changed.
• Places with more people and more roads usually have more crashes. Compare rates as well as totals.
• FARS records what happened, not why. Factors such as speeding or drinking are as recorded by police and do not establish fault.
• Comparisons in plain words, such as "about half the rate", are rounded to make them easier to read. The exact figures are shown alongside.

Privacy

FARS contains no names, and this site does not try to identify anyone involved in a crash. What the site itself collects, including cookies, analytics and email signups, is on the privacy page.

Questions about the underlying data are best answered by NHTSA's own FARS documentation and publications.