About the data

This page is the one to read before you trust anything else on this site. It says plainly what the data is, what it is not, and where it stops.

The single most important limitation

The complaint database has no denominator. It records how many people filed a complaint. It does not record how many of those vehicles were sold, registered, or are still on the road.

So a model with 1,700 complaints is not worse than a model with 300. It may simply be far more common, or older, or owned by people more inclined to file. Complaint counts on this site must not be compared between models. That is not a disclaimer bolted on afterwards — it is why this site shows the breakdown within a single model year instead of ranking models against each other.

We do not correct for sales volume, and will not invent a correction. There is no free public U.S. source for registrations or sales of sufficient quality; the usable data is commercial and licence-restricted. A rate computed from a rough guess at the denominator would look precise while being wrong, which is worse than not publishing one.

The one figure that is comparable: failure mileage

Each complaint can record the mileage at which the failure happened. That number needs no denominator: it is measured inside each model’s own reports. So “the power train is reported failing around 90,000 miles on this model year and around 50,000 on that one” is a comparison that holds, in a way that complaint counts never will.

One limit on that comparison. It works between cars of the same model year, because they have had the same number of years on the road. It does not work cleanly between a 2010 and a 2016: the newer car has simply been driven less, so its median failure mileage sits lower for that reason alone. The figure is censored from above by exposure, which is also why the window stops at 2016 — for anything newer the censoring dominates and the median stops meaning anything.

How it is computed:

• Every page shows the median and the middle half (25th to 75th percentile) of the recorded mileages, overall and per component group.
• A median is published for a component group only when at least 12 of its complaints actually carry a mileage. Below that, quantiles are noise dressed up as a statistic.
• NHTSA writes 0 into numeric fields that used to be left blank, so a mileage of zero cannot be told apart from a missing one and is counted as missing. Each page states what share of its complaints carry a mileage — typically 50–85%.

What a low median does not mean. It does not mean the car stops working at that mileage. It means the reported failures clustered early, which most often points to a fault built into a production batch — the kind of thing a warranty repair or a recall campaign addresses. Several model years on this site show medians under 10,000 miles for exactly that reason.

Complaints are not verified

Anyone can file a complaint with NHTSA. The complaints are not investigated, confirmed or technically assessed before they enter the database. A complaint records that someone reported a problem — not that the problem occurred as described, nor that it was caused by a defect. Some are duplicates. Some are later found to be unrelated to the vehicle.

This is still useful information, because volume and clustering by component are informative in aggregate. It is not evidence about any individual car.

Sources

All data comes from the U.S. federal government and is in the public domain:

• Complaints — NHTSA ODI complaint flat files (COMPLAINTS_RECEIVED_*.zip), the bulk export. It is the only source that carries the failure mileage; the JSON API does not expose that field.
• Recall campaigns — NHTSA ODI recall flat file (FLAT_RCL_POST_2010.zip).
• Crash-test ratings — NHTSA NCAP Safety Ratings API (api.nhtsa.gov/SafetyRatings), which has no flat-file equivalent.

Where the data stops

Complaints on this site run through December 31, 2024. Complaints filed after that date are not included yet.

The reason is a fault on NHTSA’s side, not a choice: the combined file that is supposed to hold every complaint since 1995 (FLAT_CMPL.zip) is currently serving a truncated export containing only 1995–1996, and the five-year instalments stop at the 2020–2024 file. The figures here are therefore complete up to the date above and will move when NHTSA’s export is whole again.

Which model years get a page

Model years 2010–2016, with at least 300 complaints on file and a recorded mileage on at least 50% of them. That is 276 pages out of roughly 5,400 model-year combinations in the window.

The window is a methodological choice. Cars newer than that have too few observations and their failure mileages are censored from above — they simply have not been driven far enough yet for the median to mean anything. There is deliberately no lower cut-off on mileage: early failures are the most informative records in the set.

The second condition exists because the mileage median is the point of these pages. A model year that cannot support one does not get a page rather than getting a thinner one.

How the numbers are computed

• A complaint, not a row. The bulk file emits one row per component and repeats the complaint’s reference number across them. Everything here counts distinct complaints; counting rows would inflate every figure on the site.
• Component groups. Each complaint names one or more components using NHTSA’s own category names, which are rolled up to the top-level group. A complaint naming two components counts once in each group, so the shares on a page add up to more than 100%. Names are shown in normal case rather than the database’s capitals; the wording is not changed. NHTSA’s category list has changed over the years, which is why a page can show both “Engine” and “Engine and engine cooling”.
• Severity shares. The share of that model year’s complaints in which the filer reported a crash, a fire, an injury or a death. One complaint can report more than one, so these do not sum to 100%.
• Bar lengths are scaled within a page. Both charts share a scale across their own rows and nothing else. They are not comparable between pages.
• Make and model names are normalised conservatively. Suffixes that denote drivetrain or duty rating are merged, because “F-250” and “F-250 SD” are one truck. Suffixes that can name a different car are not: Hyundai’s Genesis Coupe is a rear-drive sports coupe sharing nothing but a badge with the Genesis sedan, and hybrid variants are kept separate because a different powertrain fails differently.
• Checklist order. Inspection sections are ordered by how many of that model’s complaints fall in the component groups a section covers, and each section names those groups and their counts. Two sections covering the same NHTSA category — the automatic transmission and the axles both sit under “Power train” — unavoidably share the same complaints, and the order between them carries no meaning.
• Sections that do not apply are dropped. A battery-electric car gets no combustion-engine or automatic-transmission checklist: it has neither, and printing “check the oil on the dipstick” on one would be plainly wrong.

Recalls: what is and is not listed

• Only vehicle recall campaigns are shown. NHTSA files equipment recalls — tyres, child seats, aftermarket parts — under the equipment maker rather than the car, so they cannot be attributed to a model year from this file. A VIN lookup on NHTSA’s site covers everything and is the check worth doing.
• Where a campaign records the number of units it potentially affected, that figure is shown with the model years the campaign spans. It is a count for the whole campaign, often across several model years and only part of a production run. It is not a population for one model year, and nothing here divides by it.
• NHTSA’s two files do not always agree on a model’s name. The 2010 Ford Fusion Hybrid has its complaints filed under “Fusion Hybrid” and its recalls under “Fusion”, so its page shows no campaigns of its own. Its page says so rather than implying there were none.
• An absent NCAP rating means the lookup returned nothing under that model name — which also happens when NHTSA files the result under a different body or trim name. It is never a poor score.

What is not data

The inspection checklists are not computed from anything. They are written by a person — the site’s owner, who does this work on his own vehicles, including timing belts, clutches with dual-mass flywheels, injectors, axles and control arms — and they were reviewed and corrected before publication. The data decides only the order in which the checks appear on a given page.

Their wording is cautious on purpose. “Usually”, “most often” and “requires further inspection” are used instead of “means” and “is a sign of”, because a checklist that talks a buyer out of a sound car does more harm than no checklist.

No ratings, no scores, no rankings

This site publishes no reliability score, no star rating of its own and no ranking of models. There is no way to compute an honest one from data with no denominator, and a made-up number would be worse than none. The crash-test stars on each page are NHTSA’s own NCAP results, not ours.

Not advice, and not affiliated

This site is not affiliated with NHTSA, with any vehicle manufacturer, dealer or seller. Nothing here is professional or legal advice, or a substitute for a physical inspection of a specific vehicle by a qualified mechanic. Deciding whether to buy a car is yours alone.