Data study · two axes, not one list
The most reliable used cars
Lists of the most reliable used cars rank a single thing: how often something goes wrong. Your actual bill is that multiplied by what the fault costs — and across 144 models those two numbers turn out to have almost nothing to do with each other.
DVSA test recordOpen Government LicenceNo sign-up
Across 144 models compared at the same age in DVSA MOT data analysed by PriceMyRepair, how often a car fails its MOT and what the failure costs are statistically unrelated — a correlation of 0.045, which is indistinguishable from none. A Volkswagen Golf fails less often than average and costs more than average when it does. A Ford Fiesta does the exact opposite.
0.045
correlation between how often a model fails and what each failure costs — effectively zero across 144 models
4 types
of car once you plot both axes, not one ranked list — and the two interesting ones are invisible to every survey
830m
state-recorded tests behind these figures, rather than owner questionnaires or warranty claims
Reliable and cheap to fix are two different questions
In PriceMyRepair’s analysis of 144 models at ten years old, the correlation between MOT failure rate and cost per fault is 0.045 — effectively zero. Reliability rankings answer one of the two questions and quietly imply the other. The implication does not survive contact with the national test record.
The two numbers move independently. Sorting 144 models by how often they fail and comparing that with the average cost of each recorded fault gives a correlation of 0.045. Split the models into four groups by failure rate and the average cost per fault reads £114, £115, £121 and £111 — no pattern at all. Knowing how often a car fails tells you essentially nothing about what the failure will cost.
Which makes a single ranked list the wrong shape for the question. A buyer wants to know what a car will cost to keep, and that depends on both axes. Ranking one and staying silent on the other produces a list that is accurate and, on its own, not very useful.
The honest structure is a grid. Above or below the median on each axis gives four groups, and they are roughly even in size — 32 models fail often and cost more, 39 fail often and cost less, 37 fail rarely and cost more, 36 fail rarely and cost less. Two of those four are the ones no survey shows you.
The four kinds of used car
Medians across the 144 models: 28.7% of tests failed, £113 per recorded fault. Every car sits in one of these four boxes.
Fails rarely, cheap per fault — 36 models
The genuinely low-cost group, and the only one where the two axes agree. Toyota Yaris, Toyota Corolla, Honda Jazz, Mini, Skoda Fabia. Also Audi A4, which surprises people.
What every reliability list is trying to point atFails often, dear per fault — 32 models
Both axes against you. Vauxhall Vectra, Ford Ka, Renault Megane and Clio, Fiat Punto, Vauxhall Astra and Zafira.
The group worth avoiding, and the only one a single list findsFails often, cheap per fault — 39 models
Frequent visits, small bills. Ford Fiesta, Vauxhall Corsa, Ford Focus, Volkswagen Polo, Nissan Micra. Britain’s best-selling cars sit here almost to a model.
Cheap to run, and unfairly punished by survey rankingsFails rarely, dear per fault — 37 models
Long gaps, then a real bill. Volkswagen Golf, Mercedes-Benz C-Class and E-Class, Audi A3, Land Rover Freelander and Discovery — and, unexpectedly, the Fiat 500.
Reads well on any reliability list, costs more than it looksEach model is placed against the median of all 144, so “cheap” and “dear” are relative to the group rather than absolute. The full spread of cost per fault runs from £94 to £133 — narrow, which is exactly why the second axis is so easy to overlook.
Both axes in your favour
The Toyota Yaris (24.6% of tests failed, £95 per fault), Honda Jazz (21.0%, £106) and Toyota Corolla (25.6%, £94) fail rarely and cost little per fault at ten years old, according to DVSA data analysed by PriceMyRepair. Models below the median on failure rate and below it on cost per fault, ordered by how many are on the road.
| Model | Fails at ten years | Cost per fault | Tests |
|---|---|---|---|
| Toyota Yaris | 24.6% | £95 | 600,033 |
| Audi A4 | 27.8% | £103 | 537,425 |
| Mini | 27.2% | £101 | 532,187 |
| Honda Jazz | 21.0% | £106 | 379,029 |
| Skoda Fabia | 27.7% | £111 | 339,201 |
| Volkswagen Transporter | 28.1% | £113 | 294,876 |
| Toyota Corolla | 25.6% | £94 | 288,647 |
| Honda CR-V | 23.6% | £108 | 284,217 |
The Audi A4 is the row that gets argued about. It fails below the median and its faults cost below the median — a premium badge in the cheap quadrant. Cost per fault reflects what a car fails on, not what its parts cost at a dealer, and the A4 fails on ordinary wear items.
Both axes against you
The Vauxhall Vectra (36.6%, £130 per fault), Ford Ka (36.4%, £127) and Renault Megane (36.7%, £120) sit above the median on both failure rate and cost per fault in PriceMyRepair’s analysis. Above the median on both, on the same terms.
| Model | Fails at ten years | Cost per fault | Tests |
|---|---|---|---|
| Vauxhall Vectra | 36.6% | £130 | 769,798 |
| Ford Ka | 36.4% | £127 | 813,283 |
| Renault Megane | 36.7% | £120 | 679,410 |
| Renault Clio | 36.7% | £115 | 1,272,715 |
| Fiat Punto | 35.5% | £116 | 658,487 |
| Vauxhall Zafira | 32.6% | £119 | 680,454 |
| Vauxhall Astra | 32.0% | £121 | 2,112,837 |
| Volkswagen Passat | 30.7% | £120 | 538,950 |
This is the only quadrant a conventional reliability ranking reliably finds, because both axes point the same way. Everything interesting in this study is in the two quadrants where they disagree.
In PriceMyRepair’s data, the Volkswagen Golf fails 27.8% of tests at ten years old — below the median, the kind of number that puts a car near the top of a reliability list. Its faults average £115 each, above the median. It fails less often and costs more when it does.
The Ford Fiesta fails 31.5% — above the median, the kind of number that keeps a car off those lists — and its faults average £100, among the lowest here. Frequent, cheap visits.
Ranked on failure rate alone, the Golf beats the Fiesta and the story ends. Ranked on cost per fault alone, the Fiesta wins and the story ends differently. Neither ranking is wrong; each is half an answer, and the halves point opposite ways. The full failure-rate table has the first axis for all 144 models.
Why we do not multiply the two numbers together
The obvious move is to turn two axes into one figure — failure rate times cost per fault — and publish an annual running cost. We have not, and the reason is worth stating rather than hiding.
A third number is missing, and it is missing unevenly. The bill for a failed test is the cost per fault multiplied by how many faults that test recorded. That third term exists in the data but is incomplete: defect records only cover the current MOT manual scheme, while test counts cover the whole period. The shortfall is larger for models with longer histories, so any multiplied figure would quietly reintroduce a bias towards older nameplates.
So the product would look authoritative and be wrong. It would produce a clean pound figure per model, it would be quoted, and it would be measuring how long each nameplate has been sold as much as anything about the car. Two honest axes beat one misleading number, and a study that will not show its arithmetic is not worth reading.
What the MOT can and cannot tell you about reliability
This page uses the word carefully, because the test measures something narrower than the word suggests.
What it does measure
Wear and condition: brakes, tyres, suspension, steering, lamps, structure, emissions. Everything an inspector can see on a ramp in forty-five minutes.
Strong evidence, 830m tests, no self-selectionWhat it misses entirely
Gearboxes, clutches, turbos, injectors, infotainment, air-con and every electronic module. A car can pass every test it takes and still cost a fortune.
Where the four-figure bills actually liveWhy we use it anyway
Owner surveys sample people who chose to answer; warranty data covers cars under warranty. The test record covers every car on the road, whether the owner felt like filling in a form or not.
Complete population, not a sampleReliability indexes from magazines and warranty providers measure the faults this study cannot see, and are genuinely useful for that. They are answering a different question from a different population — read alongside this, not instead of it. Our guide to what the MOT actually checks sets out the boundary in detail.
Enter its reg — we’ll read the full MOT history for that exact vehicle, show every advisory a tester has recorded and price each repair for the model. A model average is a population; this is the car.
How this study was produced
Method
Source. The DVSA MOT testing bulk dataset, published under the Open Government Licence v3.0 and analysed by PriceMyRepair. 829,654,015 tests and 697,809,455 recorded defects, covering Great Britain. Northern Ireland runs a separate system under the DVA and is not in this data.
Failure rate. Tests failed divided by tests taken, for vehicles of that make and model at exactly ten years of age. Both terms are counts of tests. Models qualify at 50,000 tests or more at this age; 144 do.
Why a single age. Comparing models of different ages measures how old each nameplate’s fleet is rather than the vehicles themselves — discontinued models have old fleets and lose before anything is inspected. Every figure here is drawn from ten-year-old cars.
Cost per fault. Each recorded failure item is matched to a repair category and each category to our published independent-garage price range; the figure is the average of the range midpoints across that model’s failure items. Repair prices are our own estimates and are not part of the DVSA dataset — the two are combined here and nowhere in the official record. No vehicle-class price adjustment is applied, so the spread reflects what each model fails on rather than any assumption about what its parts cost.
Correlation. Pearson’s r between failure rate and cost per fault across all 144 models: +0.045. Quartiles of cost per fault, ordered from highest to lowest failure rate: £114, £115, £121, £111.
Why no combined figure. Set out in full above. The missing term is faults per failed test, and its incompleteness varies with how long a model has been sold.
Known limitations. One Mini appears in the record with the same make and model name and is shown here simply as Mini. Two spellings of one Toyota model exist and both were excluded rather than merged on a guess. Nameplates sold across several generations are treated as one model. Figures are not adjusted for mileage, region or use, and the tables show that use explains part of the spread. Repair categories are matched from defect text, so unusually worded faults count towards the failure rate but carry no price; coverage averages 83% of failure items. Repair ranges are estimates and final quotes come from garages.
Contains public sector information licensed under the Open Government Licence v3.0.
Most reliable used cars — FAQ
What are the most reliable used cars on MOT data?
Measured at ten years old, the Honda Jazz at 21.0% of tests failed, the Honda CR-V at 23.6% and the Toyota Yaris at 24.6% combine a low failure rate with a below-average cost per fault. That combination matters more than either number alone, because the two are independent of each other.
Does a reliable car mean a cheap car to fix?
No, and the data is unusually clear about it. Across 144 models the correlation between how often a car fails and what each failure costs is 0.045 — statistically indistinguishable from none. Roughly a third of models fail below average and cost above it, and another third do the reverse.
Why does the Volkswagen Golf appear in the expensive group?
Because it sits below the median on failure rate at 27.8% and above it on cost per fault at £115. It fails less often than most cars and costs more than most when it does. Neither number is bad; together they describe a car with long gaps between real bills rather than frequent small ones.
Is the Ford Fiesta expensive to run?
Not on this measure. It fails more often than the median at 31.5% but its faults are among the cheapest at around £100 each. Frequent, small bills rather than rare, large ones — a profile that reads badly on any ranking of failure rates and rather well on a bank statement.
Why not just multiply failure rate by repair cost?
Because a third number is needed and it is incomplete. The bill for a failed test is the cost per fault multiplied by how many faults were recorded, and defect records cover a shorter period than test counts. The shortfall is larger for long-established models, so a combined figure would look precise and reintroduce a bias towards older nameplates.
How is this different from magazine reliability surveys?
Those sample owners who chose to respond and mostly cover cars in their first years. This covers every test taken by every car on the road, including the ones nobody would fill in a form about. It also measures different faults — wear items rather than gearboxes and electronics — so the two are complementary rather than competing.
Do premium cars cost more per fault?
Some do and some clearly do not. Mercedes-Benz E-Class faults average £134 while Audi A4 faults average £103, below the median for all models. What drives the figure is what a car fails on, not what its badge implies — and no price adjustment for vehicle class was applied to these numbers.
Which used cars should I avoid on this data?
The group above the median on both axes: Vauxhall Vectra, Ford Ka, Renault Megane and Clio, Fiat Punto, Vauxhall Astra and Zafira. All are older nameplates, and part of what the figures capture is how such cars are typically used and maintained rather than how they were built.
Does a good MOT record mean a reliable car?
It means a car that wears well in the areas the test inspects. Gearboxes, clutches, turbos and electronics are outside the test entirely, and that is where a large share of expensive breakdowns sit. A clean history is real evidence, just narrower evidence than the word reliable usually implies.
How do I check a specific used car before buying?
Its MOT history is public and names every advisory a tester has recorded, with dates. That beats any model average, because repeated advisories on the same corner are next year’s repair announcing itself. Our used car check reads that history and prices the work from the registration alone.
Related studies and guides
The same dataset cut along the other axes.