How CarWhere rates model years
Every year-by-year verdict on this site is computed from federal data, not opinion. We read the government’s own complaint and recall records, weight them for severity, and compare each model year against the rest of its own model. Here is exactly how — and where the method’s limits are.
By Sam Reynolds, Lead Researcher · Updated 2026-07-14
What each verdict means
Avoid. This model year carries a materially elevated federal complaint burden versus the rest of the model — often a specific, well-documented failure. Buy only with eyes open and a thorough inspection.
Caution. Above the model’s own average, or carrying an open concern worth checking, but short of the avoid line.
Solid. A middle-of-the-pack year for this model — no standout federal red flags.
Best. Among the cleanest federal records for this model, with enough years on the road to trust the signal.
Insufficient data. Too few complaints on file to rate fairly — we never call a year “best” simply because it is too new or too rare to have generated reports yet.
Where the data comes from
- • NHTSA owner complaints (Office of Defects Investigation) — every complaint a vehicle owner has filed with the federal government, including the component and the owner’s own account.
- • NHTSA recall campaigns — the safety recalls issued for each model year and what the defect can cause.
- • Owner narratives — the free-text of those federal complaints, which we synthesize into named issues and cross-check against independent owner discussion.
- • CarWhere verified buyer prices — what real buyers actually paid, used to point you at the good years, not just away from the bad ones.
How the score is built
- 1
Normalize the vehicle. Resolve the make and model to NHTSA’s canonical names (vPIC) and enumerate every model year on the road.
- 2
Pull the federal record. For each model year, collect all NHTSA owner complaints and every recall campaign, cached and refreshed monthly.
- 3
Weight for severity. Complaints that report a crash, fire, injury, or death — or that name a safety-critical system such as the engine, transmission, steering, brakes, or fuel system — count more heavily than a rattle or a trim complaint.
- 4
Normalize within the model. Annualize complaints by the model year’s time on the road — this partially adjusts for exposure time (older years have had longer to accumulate reports) but does not adjust for how many vehicles were sold. Then compare each year against the model’s own median, so a verdict means “worse than a typical year of this model,” not “worse than a Corolla.”
- 5
Fold in recalls and confirmed issues. Add weight for recall campaigns — more for those describing fire or a crash mechanism — and for issues our synthesis step has independently confirmed.
- 6
Apply guardrails, then verdict. Years with too few complaints are marked “insufficient,” the two newest model years are never rated “best,” and a manual override can correct any year. The final score maps to avoid / caution / solid / best.
How confident we are in each named issue
When we describe a specific failure — a transmission that judders, an engine that burns oil — we label how well the evidence holds up:
- • Confirmed — a federal complaint cluster and independent owner reports agree, or a recall backs it. Published with the numbers.
- • Reported — owners describe it, but the federal signal is thin. Published with a hedge.
- • Myth-check — forum lore the data contradicts (for example, complaints that fell sharply after a fix). We say so.
Limitations, stated plainly
- • Complaint counts are raw federal filings, not adjusted for how many of each model year were sold. A popular year attracts more complaints simply by being on more roads; we normalize within a model to soften this, but no public dataset gives exact sales by model year.
- • Newer model years have had less time on the road, so their record is thinner and read with more caution.
- • A verdict is a screen, not a substitute for a pre-purchase inspection of the specific vehicle and VIN.
How often it updates
Complaint and recall counts refresh monthly — the reason these pages don’t go stale the way static “worst years” articles do — and the owner-narrative synthesis is re-run quarterly. Every page shows the month its data was last refreshed.