Methodology

We use four NHTSA Office of Defects Investigation (ODI) datasets, refreshed on a monthly cadence:

• Consumer complaints — every complaint filed with NHTSA since 1995, including the free-text narrative, component, and crash/fire/injury/death flags.
• Recall campaigns — every safety recall since 1967, with the defect, consequence, and remedy summaries and the number of units affected.
• Defect investigations — ODI investigation actions, including whether each is still open.
• Safety ratings — NCAP crash-test star ratings, where available (2011 onward).

Make and model names in the complaint files are inconsistent, so we normalize them against NHTSA's own vPIC canonical vehicle catalog before aggregating.

A model year's complaint count is the number of distinct NHTSA complaints whose make, model, and model year map to that vehicle. Recall counts are distinct recall campaign numbers affecting that model year. We do not add, weight, or estimate any figures — every count is a direct tally of government records.

A model year is flagged as one to avoid when either condition holds:

• Its complaint count is at least 1.5× the median year for that model and it has at least 30 complaints (so a single noisy low-volume year can't trip the rule); or
• It has an open NHTSA defect investigation.

The “best year” is the model year with the fewest complaints among years that are not flagged, provided the model has at least three years of data to compare.

NHTSA labels each complaint with a detailed component code. We group these into about 20 human-readable categories (engine, transmission, brakes, electrical, airbags, and so on). The year-over-year trend for a component compares its complaint count to the prior model year (up if more than 25% higher, down if more than 25% lower).

The short quotes on model-year pages are real complaint narratives, trimmed to about 300 characters. Before display we automatically remove vehicle identification numbers, phone numbers, and email addresses. These are consumer statements filed with NHTSA — they are not verified defects, and their accuracy is the responsibility of the person who filed them.

• Complaints are self-reported and unverified. Anyone can file one; NHTSA does not confirm the underlying defect.
• Counts are not normalized by sales. There is no public per-model-year sales data, so a best-selling vehicle naturally accumulates more complaints than a rare one. Compare years within a model, not raw counts across models.
• Reporting lags. Recent model years have had less time to accumulate complaints, and problems often surface years after sale.
• No repair costs. NHTSA data contains no cost information, so we publish none.

We deliberately publish no invented reliability score beyond the transparent, rule-based metrics described above. If a section would require data we don't have, we omit it.

The underlying NHTSA files are refreshed monthly. Each page shows the date its data runs through. Source documentation lives at nhtsa.gov/nhtsa-datasets-and-apis.

Questions about the data? See About.