How these numbers are calculated

Everything here is derived from public federal data. The method matters, because the obvious way to read this data is wrong in three specific ways.

Sources

Complaints — NHTSA Office of Defects Investigation consumer complaint file, every safety complaint filed since January 1995. 2,175,102 of them concern vehicles.
Recalls — NHTSA’s recall campaign database, queried per model year.
Both are US federal public-domain records. This site is not affiliated with NHTSA.

The headline figure

For each vehicle we compute the age at failure of every complaint — how far into the vehicle’s life the reported incident happened — and then the share of complaints that arrived in the first year of service. A high share points to problems present from new; a low share points to ordinary wear. The spread across vehicles is wide and genuinely discriminating, running from under 1% to over 90%.

Three corrections

Mileage could not be the primary measure. NHTSA changed systems in May 2021 and mileage capture fell from roughly 83% of complaints to 13–18%. The incident date survived at 100%, so age at failure is the primary axis and mileage corroborates it where present. Across 1,186 model years with enough data on both, the two agree closely (r = 0.74).
New cars cannot have old failures. A 2026 vehicle physically cannot have a five-year-old failure, so its first-year share is 100% by construction. Every ranking therefore compares a vehicle only against others of the same model year. Comparing first-year share across model years is meaningless and is never done here.
Media storms look like defects. In February 2010, during the unintended-acceleration coverage, Toyota logged 11× its normal complaint volume across 33 different models in a single month. Left in, that brands every 2010 Toyota a lemon. Fourteen such make-months are excluded, identified as months where one manufacturer’s volume far exceeds its local baseline across many models at once.

Where a claim is withheld

A verdict is a statement about an identifiable product, so it is only published when the data can carry it. Two tests have to pass.

The margin of error has to separate the vehicle from a typical one. First-year share is a proportion, and on 30 complaints its 95% interval spans roughly ±16 points — wide enough that a striking figure can be noise. A vehicle is only called an early-defect pattern when the lower bound of that interval still clears the 90th percentile of its own model-year cohort. 960 model years reach a rank extreme but fail this test; they are shown with their numbers and no verdict. The test is self-calibrating rather than a flat sample floor: a vehicle at 80% first-year complaints clears it on less data than one at 32%, because the claim is that much stronger.
The model year has to be old enough to measure. For a very recent vehicle nearly every complaint is a first-year complaint, because that is all the service life there has been. Across MY2026 the median is 100% by construction. Model years whose cohort median passes 30% are marked too new to assess and carry no verdict.

The difference between marques

Averaged across their model years, marques do not sit evenly. Subaru averages a rank of 69 and MINI 28, on a scale where 50 is neutral; the spread across the 33 marques with substantial records has a standard deviation of about 7 rank points.

We tested whether this is an artefact of sales volume and it is not: the correlation between complaint count and first-year share is −0.12 overall and effectively zero within any single model-year cohort. So the difference is real and persistent. What we cannot tell you is what causes it. Genuine engineering differences, differences in how likely a marque's owners are to escalate to a federal complaint, dealer networks, and warranty terms would all produce the same signal, and nothing in this data separates them. Treat comparisons between marques with that firmly in mind.

What these numbers cannot tell you

This is not a reliability rating. Complaints are self-selected: people file about failures that frightened or stranded them, not about routine wear. A vehicle with more complaints may simply have sold in greater numbers.
Complaint counts are not normalised by units sold, because NHTSA does not publish sales volumes. Comparisons between two different vehicles should be read with that in mind.
Owner populations differ. Some brands’ owners are more likely to file federal complaints than others, and nothing here corrects for that.
A listed recall may already have been done. Recalls are issued against a model year, not a specific car. Check an individual VIN at nhtsa.gov/recalls.
Minimum sample sizes are enforced: a model year needs at least 30 complaints to appear at all, an individual system at least 15, and a model-year cohort at least 20 members before any rank is published.

Not safety advice

This site summarises public records for general information. It is not advice on whether a particular vehicle is safe to buy or drive. For anything safety-critical, consult the manufacturer, a qualified mechanic, or NHTSA directly.