How to read NHTSA complaints before buying a used car

Owner complaints are public, free and the best early warning there is for engine and transmission trouble. Here is how to read them without being misled by the raw count.

The Lane Check · September 14, 2026

Every complaint an owner files with the National Highway Traffic Safety Administration is public. Anyone can search them by year, make and model, for free, and they go back decades. It is the largest body of first-hand reports about what goes wrong with cars in the United States, and almost nobody buying a used car looks at it.

Part of the reason is that the raw data is hard to read well. This guide is how we read it, and how you can do the same on your own.

What a complaint is

A complaint is a report an owner chose to file with NHTSA's Office of Defects Investigation. It is not a warranty claim and not a repair record. Each one carries:

• The vehicle: year, make, model, and usually the VIN.
• A component category, picked by the owner or by the NHTSA agent who took the call: Power Train, Engine, Fuel/Propulsion System, Electrical System, Air Bags, Steering, and so on.
• A date the incident happened, and the date it was filed.
• Flags for whether there was a crash, a fire, an injury or a death.
• A summary in the owner's own words. Some are a sentence. Some are a page.

Summaries that begin with something like TL* THE CONTACT OWNS A 2013 FORD FOCUS were taken over the phone and typed up by NHTSA staff, which is why they read like a police report. The rest were typed by the owner.

Rule one: read the share, not the count

The single biggest mistake is to be impressed or reassured by the number of complaints. Popular cars collect more complaints because there are more of them on the road, and older cars collect more because they have had longer to break.

The 2010 Toyota Corolla has 1,264 complaints on file. The 2019 Corolla has 202. On the count alone the 2010 looks six times worse. But only 11% of the 2010's complaints are about the engine or transmission, against 64% for the 2019. The 2010 has a lot of complaints about a lot of things, which is what a very common ten-year-old car looks like. The 2019 has a concentration, which is what a problem looks like.

So the number we use is the drivetrain share: complaints tagged Power Train, Engine or Fuel/Propulsion, divided by all complaints for that model year. Volume cancels out. What remains is where the trouble is concentrated.

Rule two: know what the categories mean

Power Train is the transmission and everything between the engine and the wheels: automatic and CVT failures, shudder, slipping, driveline, axles, transfer cases. Engine covers the engine itself, oil consumption, stalling, timing components. Fuel/Propulsion System is where hybrid drive and direct-injection problems often land. We count all three, because owners are inconsistent about which one they pick, and a stall from a fuel-delivery fault is a drivetrain problem to the person it happened to.

The rest is context. Electrical System is a catch-all that includes infotainment, so a high count there usually means annoyance rather than danger. Air Bags and Seat Belts spike when a recall campaign is in the news, which brings us to the next rule.

Rule three: watch for recall flooding

When a big safety recall hits, owners file thousands of complaints about it, all in one category. The Takata airbag recall did this to a whole generation of cars. Those complaints are real, but they swamp the denominator: a car with a genuine transmission problem can look fine on the share because 40% of its complaints are suddenly about airbags.

Our scoring handles this by computing the share a second time with the flooded category removed, and keeping whichever signal is stronger. On a public car page you will see a note when that adjustment was applied. If you are reading the raw data yourself, look at the component counts before you trust the ratio.

Rule four: look at the years either side

One bad year in a run of good ones might be a blip, a single fleet operator filing in bulk, or a genuine one-year defect. A run of bad years is a design. The Subaru Outback is a clean example: 2011 through 2014 sit between 41% and 62%, then 2015 drops to 18% and stays low.

That shape tells you two things. The problem was real, because it persisted. And it was fixed, because it stopped. If you are looking at a 2014 and a 2015 side by side, the data is telling you which one to buy.

Rule five: read the summaries for a repeated symptom

Once the share says there is something to look at, the summaries tell you what it is. Do not read for drama. Read for repetition. Twenty different owners describing the same thing in their own words is the signal: a shudder between 20 and 40 mph, a hesitation pulling away from a stop, a gas smell in the oil, the engine stalling at low speed with no warning light.

That symptom is what you take into the test drive and what you ask the seller about. A history report cannot give you it. A mechanic doing a general inspection will not go looking for it unless you tell them to.

Rule six: respect the sample size

Under about 25 complaints, the share is noise. Three complaints, two of them about the transmission, is 67% and means nothing. We do not score a model year until it clears that floor, and the most recent two or three model years often have not. That is not a clean bill of health. It is too early to tell.

The same caution applies at the other end. Complaints are self-selected. People file when something went wrong, so every car's file skews negative, and a share of 20% does not mean a fifth of those cars have drivetrain trouble. It means that among owners unhappy enough to file, drivetrain was not where the unhappiness concentrated.

Doing it yourself

NHTSA's search is at nhtsa.gov, under Recalls, then search by year, make and model, then the Complaints tab. For a popular car expect several hundred to a few thousand entries. To read them the way described here you would tally the categories, compute the share, repeat for the neighbouring years, and skim a few hundred summaries for the repeated symptom. It takes an evening per car.

That evening is what the tool does in a few seconds, on the same public data, and it is why the first report is free. But the method is not a secret, and if you would rather do it by hand, now you know how.