How We Score Used Car Reliability

Each part answers a different question, and none of them tells the whole story alone. A car can have few complaints but many recalls. An older model can look fine over its life and still be failing now. So we look at all four.

Serious problems carry the most weight because they can be dangerous and cost the most to fix. The last two years come next, because they show how cars of that year are holding up today, and that is the car you would buy. Recalls show safety defects the maker confirmed. Everyday problems matter too, but a noisy radio is not a failed transmission, so they count the least.

1. Count what owners report. Every complaint filed with NHTSA for the model year. Complaints that only say the owner is waiting for a recall part, with no fault, are left out. A fixed set of rules sorts each complaint as serious or everyday, from its words and from the crash, fire, injury and tow flags NHTSA records. The same rules sort the complaints of the average car, so every car is measured the same way.
2. Adjust for how many were made. A popular car collects more complaints just because there are more of them. We divide by the number of cars the automaker reports making for the U.S. for that exact model year.
3. Compare with the average car of the same age. Older cars collect more complaints. Each year is measured against the average car at the same age, and in the same calendar years, so a 2012 and a 2022 are judged fairly.
4. Turn each comparison into points. Each part is scored on its own. Matching the average car scores 70 on that part, and every doubling costs that part 10 points: twice the average scores 60, half the average scores 80, and values in between fall in between. When the counts are small, as they usually are for recalls, we pull the comparison toward the average, so one or two recalls cannot swing the score far.

5. Weigh the parts and add them up. Serious and everyday problems are combined into one comparison first, with serious problems weighted 35 and everyday problems 15. That way one more serious complaint always lowers the score more than one more everyday complaint. Together they count for half of the score. The last two years count for 30% and recalls for 20%. So if serious and everyday problems both double, the score drops 5 points. Doubling the last two years costs 3, and doubling recalls costs 2. When a year skips the last two years, serious and everyday problems count for 80%. The parts do not need to agree: a year can score 60 on one part and 85 on another, and its score is the weighted average.

The 2013 Honda CR-V scores 85. Every model year starts at 100. Each part then loses its weight times the points it falls short of 100. Recalls, for example, score 78.2, which is 21.8 short of 100, and 20% of that is 4.4.

On our pages the score is rounded to the nearest whole number, 85. Each model-year page shows the same points lost as whole numbers, with serious and everyday problems on separate lines.

• Excellent. Score 80 or higher. About half the problems of the average car its age, or fewer.
• Good. Score 70 to 79. Up to an average number of problems for its age.
• Caution. Score 60 to 69. Up to twice the problems of the average car its age.
• Avoid. Score below 60. More than twice the problems of the average car its age.

Scores are rounded to the nearest whole number, and the rating follows the rounded score. A year at 79.5 shows as 80 and is rated Excellent. A year at 79.4 shows as 79 and is rated Good.

Small gaps between scores don't mean much. Complaint counts go up and down by chance, so two scores less than 3.5 points apart are too close to call. We found that number by scoring many random re-draws of the same complaints and measuring how far the scores moved.

A final score needs 36 months of complaints. We count a model year's months from July of the year before its name, because model years usually go on sale then. A 2024 model, for example, starts in July 2023 and reached 36 months in July 2026, so it has a final score. Before 36 months, we show what can be said honestly.

• Up to about 21 months. Complaints, recalls and investigations so far, the scores of earlier years from the same generation, and when a provisional score is expected. The production report for the year is not complete yet, so no rate is calculated.
• About 21 to 24 months. An early signal when complaints already run far above or below average.
• 24 to 36 months. A provisional score with the range the final score is likely to fall in.
• 36 months and later. The final score and rating.

NHTSA files each complaint under broad categories such as engine or power train. To say which part failed and what happened, we read every complaint one by one and record the part, how it failed and the outcome, with a quote from the complaint. Then we group the complaints by problem and count them. Where we have not finished reading a model year yet, its page says how many complaints we have read so far. When NHTSA publishes new complaints, we read the new ones and add them. This reading names the problems on each page. It does not change the score, which uses the rules above.

• Complaints are reports filed by owners. They are not verified defects, and many problems are never reported.
• Reporting habits change over time and with news coverage. Comparing with the same calendar years reduces this effect but does not remove it.
• A score describes a model year, not one car. A well-kept car from a low-scoring year can be fine, and the reverse.
• Commercial vans and fleet vehicles are left out, because fleets rarely file complaints.