Methodology

This site is in preview. Car licensing figures (models, versions, makes, rankings and the national count) are live DfT and DVLA data, MOT figures are live DVSA data, and road traffic and road safety figures are live DfT data, as are local figures (vehicles, electric cars, chargers and population by area), driving test pass rates, engine sizes (DfT), official fuel economy (VCA) and safety recalls (DVSA). Other figures are demo data until their datasets are connected; the methods below are the ones they will use. We will record any change to a method here with its date.

Principles

• Every figure shows its source, the period it covers and when we last updated it.
• We publish official figures as released. Where we calculate something new, we say so and explain how.
• We do not publish a page unless there is enough data to make it genuinely useful.
• We never estimate a figure the source does not support, and we do not fill gaps with guesses.

Licensing counts

Counts of licensed and SORN vehicles by make and model come from the Department for Transport’s vehicle licensing data files (df_VEH0120_GB), which are based on DVLA records. We group models by DVLA’s generic model name, so every generation of a model counts together. Changes compare the same quarter a year earlier, which removes seasonal effects. First registrations come from df_VEH0160_GB and include used imports registered in Great Britain for the first time.

Since about 2023 DVLA has recorded many new cars with a trim level in the generic model field (“Seal Design EV”, “5 Iconic E-Tech EV”), which would split one model across several names. We fold these into their model, using a table we check against each release. Where a new car shares its name with a classic, it keeps a distinguishing word (the electric Ford Capri is “Capri EV”, the new Renault 5 is “5 E-Tech”), so the classic’s figures are not mixed with the new car’s.

These files cover Great Britain (England, Scotland and Wales) and go back to 1994, which is what makes long “how many are left” histories possible. The UK-wide files, which add Northern Ireland, only start in 2014. Every page says which area its figures cover.

Year-by-year charts use the end-of-year count; the latest point is the most recent quarter. Where a model’s highest count is the first year of the data, we say so rather than calling it a peak, because the model may have been more common before the data begins. We only show a registration history when a model has had at least 100 first registrations in a year. Catch-all groups such as “Other British” are not treated as makes.

Versions of a model

DVLA also records a more detailed name for each car, such as “205 GTI” or “ESCORT RS 2000”. These names are too many and too inconsistent to use as they are: the same car can appear as “GOLF GTI”, “GOLF GTI S-A” and “GOLF GTI (115 BHP)”. We group them into versions by setting aside details that do not make a different version: engine size, valves, power, gearbox, number of doors, fuel, and engine badges such as TSI. Words that name a version stay, so the 205 GTI, Rallye and CTI are separate versions, and so are the Fiesta ST and the Fiesta ST-Line. A short, hand-checked list joins names the rules cannot know are the same car, such as “ESCORT COSWORTH” and “ESCORT RS COSWORTH”. Every version page lists the DVLA names it is built from. We never create a version that is not in the data.

A version only gets its own page when it differs from its model in a way the data can show. We compare the share of its cars still registered (licensed or SORN) with other versions of the same model from the same years. Enthusiast versions stand out: 12.7% of Peugeot 205 GTIs are still registered, against 2.2% of other versions of the Peugeot 205 of the same age. For versions mostly first registered since 2001, “the same years” means the same average year of first registration. For older versions, it means their licensed count peaked within a few years of each other, looking only at years up to 2001. Versions sold both before and after 2001 have no single age, so we do not compare them.

A version needs at least 100 cars (licensed or SORN) and 1.5 times the survival rate of its peers, and at least one percentage point more, to get a page. To be indexed by search engines it needs twice the rate, 150 cars and 5 years of history. A few well-known versions that are still on sale or were sold for decades, such as the Golf GTI and the Fiesta ST, cannot show a difference this way, so we list them by hand. They still need the same amount of data.

MOT pass rates

MOT figures come from DVSA’s anonymised record of every MOT test in Great Britain, and cover the latest complete calendar year. We use normal tests of class 4 vehicles (cars and light vans). A first-time pass is a normal test passed outright. A fail, or a pass after defects were fixed at the test centre within the hour (PRS), counts as a first-time failure. We exclude retests and abandoned or aborted tests.

Age is the time from a car’s first use to the test, rounded to the nearest year, so a car tested a few weeks before its third birthday counts as three. We only use cars aged three and over, as newer cars do not normally need an MOT.

DVSA records the make and model typed in at the test. We match those names to DVLA’s generic models, using the model’s own name, then known variants (a BMW “320D” is a 3 Series), then the shortest DVLA name it starts with. “All cars” means every test matched to a car model (98% of class 4 tests in 2025). Because class 4 also covers light vans, we leave out any model where more than 15% of those on the DVLA record are vans (such as the Citroën Berlingo and Land Rover Defender), and any model with far more tests than cars on record, which signals a poor name match. Those pages say why they have no MOT figures.

Older cars fail more often, so raw pass rates mostly reflect a model’s age. We compare each model with all cars of the same age at test, in age bands (3, 4–5, 6–7, 8–10, 11–14, 15–19, and 20 years and over). The “same ages” figure is what all cars would score if they had the model’s mix of ages. A model needs at least 1,000 tests before it gets MOT figures, and at least 500 in a band before we show that band.

Failure categories follow the sections of the MOT inspection manual. A failed test counts once in each category it has a failure in, however many defects it has there, so shares add up to more than 100%. We show categories once a model has at least 300 failed tests. Dangerous defects are those the tester marked as dangerous.

Annual mileage comes from consecutive normal tests of the same vehicle between 300 and 430 days apart: the change in odometer reading divided by the time between the tests. We discard pairs where the reading went down or implies more than 60,000 miles a year. A model needs 500 such pairs for an average, and 200 in an age band.

Model years

A model-year page, such as the 2014 Ford Fiesta, covers the cars of one model first registered in one year. The year is DVSA’s first-use date: when the car was first registered, in the UK or abroad. Each page compares the cars with every car first registered the same year, which puts like against like: all of them are the same age.

MOT figures come from the latest complete year of DVSA’s anonymised MOT results and follow the method above (normal tests, cars three and over, retests left out, the same name matching and van screening). Years of first registration after the third year before the MOT year are left out, because those cars had not had an MOT. The previous year’s pass rate is the same cars tested a year earlier.

How many are left comes from DfT table VEH0124: cars licensed or declared SORN at the end of each year from 2014, by model and year of first registration. It covers the United Kingdom, including Northern Ireland, while MOT figures cover Great Britain; each page says which is which. For cars first registered before 2014 the count starts at the end of 2014. Cars scrapped, exported or otherwise off the record leave the count; late registrations and imports can join it, so the number can rise in the first years.

Mileage is the odometer reading at each car’s latest normal MOT of the year, one reading per car, between 1 and 999,998 miles. We show percentiles, rounded to the nearest 100 miles: the median (half of cars had covered less), the middle half (25th to 75th percentile) and the middle 80% (10th to 90th). For example, half of the 2014 Ford Fiestas tested in 2025 had covered fewer than 70,800 miles, and a quarter more than 91,200. Testers type the reading in, so a few are in kilometres or mistyped; percentiles are not moved by a handful of outliers.

A model year needs at least 500 MOT tests to get a page, and 1,000 to be indexed by search engines, with at least 7 measures and 3 years of history. With 500 tests a pass rate is accurate to about four percentage points either way; with 1,000, about three. Its model must have a page too. Lists of defects need at least 150 failed tests, and each defect or advisory listed at least 10 tests; mileage needs 300 cars.

Specific defects and advisories

Besides the category, DVSA records the exact reason for each defect, such as “fractured or broken” for a coil spring. On their own these descriptions are often unclear, so we put the part in front (“Coil spring: fractured or broken”), use the parent part where the part’s name is generic (“Condition”) or repeated in the description, and write the result in sentence case. DVSA also records where on the car a defect was (nearside, front and so on); we ignore that, so the same fault on any wheel or side is one line, and identical descriptions are combined.

A failed test counts once for each defect it has, however many times the defect was recorded. Defect shares are of failed tests (fails and passes after repair at the station). Advisories are things the tester noted that did not fail the test; their shares are of all tests, and we use DVSA’s advisory wording. A model needs 1,000 tests and 150 failed tests for its lists.

Reliability ranking

The most reliable cars ranking covers every model with MOT figures (at least 1,000 tests). Each model’s first-time pass rate is compared with what all cars would score with the same mix of ages, using the seven age bands above and weighting each band by the model’s own tests. The difference, in percentage points, is the score. Makes combine all their models, with at least 5,000 tests; fuel types use the fuel DVSA records for each car tested. Classes (supermini, family SUV and so on) are our own labels for the most popular models, not DVLA’s.

The score measures how cars do at the MOT, which depends on how they are driven and looked after as well as how they are built. Rarely driven sports and luxury cars do well; models often used as taxis or on high mileages can do worse. With 1,000 tests a pass rate is accurate to about three points either way, so small differences between rare models mean little.

MOT results by postcode area

DVSA records the postcode area of the test station, not of the car’s keeper. We group tests by that area (the letters at the start of a postcode, such as B for Birmingham) and name each area from Royal Mail’s list; an area usually covers much more than its main town. Tests with a missing or unknown area count towards Great Britain but no area. Each area is compared with what all cars in Great Britain would score with its mix of ages. Annual mileage uses consecutive MOT readings of the same car, as above, placed in the area of the later test. An area needs at least 1,000 tests for a page; every area in Great Britain has far more.

Car colours

The colour of each car MOT tested in the latest year, counted once however many tests it had, grouped by year of first registration. This shows the cars still on the road, not every car sold, so for older years it describes the survivors. We show the unbroken run of recent years with at least 10,000 cars each. Colour is the broad colour on the car’s record, so every shade of grey counts as grey.

Official fuel economy and CO2

Fuel economy, CO2 and Euro standards come from the Vehicle Certification Agency’s car fuel data: snapshots of the new cars VCA lists, from July 2000 to 15 September 2026. They are official type-approval figures from laboratory tests. VCA says they may not be achieved in real-world driving, that they should not be used to rank cars with similar figures, and that other pollutants should only be compared between cars of the same Euro standard. We publish combined mpg, CO2 and the Euro standard, and no other pollutant figures.

Snapshots to September 2018 carry figures from the NEDC test, and those from the 2019 guide figures from the WLTP test. The two tests give different results, so we keep them apart: model pages show them in separate tables, and the CO2 trend is two series that are never joined. We use VCA’s combined figures; for plug-in hybrids, VCA’s weighted figures. We check every row: a petrol or diesel car whose CO2 and mpg do not agree (by the fixed relationship between fuel burned and CO2, within 20%) is left out as a data error, as are rows with no combined figure.

VCA’s manufacturer and model names are matched to DVLA’s generic models in the same way as MOT names, after removing model years, chassis codes and body words (“Evoque, 20MY” is a Range Rover Evoque; a MINI “3-door Hatchback” is matched on its description, such as Cooper S). A match is dropped if DVLA recorded fewer than 25 new registrations of that model in the years VCA listed it, which catches names shared with an unrelated older car: the electric Renault 5 of 2025 is not the original Renault 5. Versions are matched only to versions that already have a page, using the same rules as DVLA’s names; well-known versions such as the Golf GTI also match when their name appears anywhere in VCA’s description.

On model pages, figures are grouped by fuel and engine size (to the nearest 0.1 litre), with the years VCA listed them. A version is one line in VCA’s data (description, gearbox, engine and fuel), counted once however many snapshots list it. Rankings are in bands (5 mpg, or 0.5 miles per kWh for electric cars) with cars in alphabetical order inside each band, so no car is put above another on a fraction of a mile per gallon. Average CO2 by year uses the last snapshot of each year and counts each version once, so it describes the cars on offer, not the cars sold. VCA’s data is Crown copyright; its reuse licence is being confirmed with VCA.

Not every snapshot lists every manufacturer. We compare each snapshot with the last NEDC one, which lists every major make, and treat a make as missing when it has almost no versions in a snapshot although VCA lists it in other years. VCA’s data for 2019, 2020, 2021, 2022, 2023, 2024, 2026 leaves out makes such as Audi, BMW, Ford, Infiniti, Jaguar, Land Rover. We publish average CO2 only for years with every major make (2000 to 2025), the economy rankings say which makes the latest file leaves out, and model pages say when a model’s make is missing from some years.

Engine sizes

Engine sizes come from DfT table VEH0220: licensed cars at the end of each year from 2014, by make, model, fuel and engine size in 100cc bands (for example 1,501cc to 1,600cc), based on DVLA records. The table covers the United Kingdom, including Northern Ireland, so its totals are a little higher than the Great Britain counts on the rest of a model page; we say so wherever both appear. SORN vehicles are excluded. Electric cars have no engine size.

Models and versions are matched on the same DVLA names as the licensing counts. On a model page we show sizes that reach 2% of the model’s licensed cars in any year (smaller ones are grouped as other sizes), and changes in share once the model has at least 100 licensed cars. We mention a change in the share of an engine size when it is at least five percentage points.

Road traffic

Traffic figures come from DfT’s annual average daily flow (AADF) at count points: the average number of vehicles passing a point on a typical day, both directions combined, by vehicle type. Every link of a motorway or A-road has a count point and a figure every year. In a given year DfT counts some links and estimates the rest from earlier counts and local trends; each road page says how many were counted.

A road is every link with its number, in England, Scotland and Wales. A1 and A1(M) are separate roads. Vehicle miles are AADF multiplied by link length and 365, summed over the road. Averages along a road or a section are weighted by link length. We build pages for motorways with at least two count points and A-roads with at least five, with ten or more years of figures.

Sections: DfT’s major road database draws each count point link as a line between two junctions. We join the links into the road’s route, set aside short links off the main line (spurs and junction link roads, which still count in the totals), and merge links split only at a council boundary, so each section runs from junction to junction. Route maps are drawn from the same lines and are schematic.

Traffic change is like for like: we compare only count points with a figure in both years, weighted by today’s link length, so a new bypass or a renumbered stretch does not look like growth. The yearly charts show the whole road as it was each year. Rankings of growth need 10 miles of road and 70% of it comparable. Council area traffic on all roads, minor roads included, is DfT’s own estimate for each highway authority; minor-road traffic for Great Britain is that total less motorways and A-roads.

Collisions

Collisions come from STATS19, the police record of personal injury collisions on public roads in Great Britain, for the last five years (2021 to 2025). Damage-only collisions are not included, and some injury collisions, particularly slight ones, are never reported. A death counts as fatal if it happened within 30 days of the collision.

Severity: police forces have been moving to injury-based reporting systems, which record more injuries as serious. DfT publishes adjusted serious and slight figures that estimate what every force would have recorded that way, and we use them for every serious and slight figure, so years and areas compare fairly. “Fatal or serious” (killed or seriously injured) is fatal plus adjusted serious.

Roads: a collision belongs to the road police record as its first road (motorway, A(M) or A-road, by number). Collisions at a junction where the road is the second road are counted separately and left out of rates. STATS19 records the M6 Toll under the M6’s number, so its collisions are counted on the M6. Collisions with a recorded position are placed on the nearest section of their road (within 2 km).

Rates: collisions per 100 million vehicle miles divide collisions over the five years by DfT’s vehicle miles for the same road and years. We show a rate for a road with at least 20 collisions and a section of half a mile or more with at least 10. Rankings of roads need 10 miles and 50 collisions. Council area rates use DfT’s traffic estimate for all roads in the area, so they are given for highway authorities (counties, unitary authorities, metropolitan and London boroughs, Welsh and Scottish councils) but not for districts.

Privacy: we publish counts only, never a single collision or casualty. Below Great Britain level, a count of one or two is shown as “fewer than 3”. Where a hidden count could be worked out from the others (a total with one year hidden, or serious injuries next to a hidden fatal count), we hide that figure too.

Fuel prices

Fuel prices come from the statutory Fuel Finder scheme. We use each station’s latest price, give every station equal weight, and leave out stations that have not updated their price in seven days. We publish averages and ranges, not the raw price feed.

Roadworks and road closures

Two registers publish this, and they do not cover the same thing, so we keep them apart. For England we use National Highways’ Road and Lane Closures service: planned closures on the motorways and trunk A-roads it manages, not council roads. For Scotland we use the Scottish Road Works Register, which covers every road down to a residential street.

England: we read the next fortnight twice a day and publish planned closures only. Closures caused by incidents are in the same service, but they change minute by minute and these pages are built a few times a day, so publishing them would tell you a road is shut when it has reopened. We count one entry per scheme, which is how the service groups a job: a scheme can cover several nights, directions and stretches of road, so counting its individual records would multiply the same closure many times over. A scheme’s period runs from its earliest start to its latest end, and the times shown are London time. Schemes the service marks suspended have been cancelled, so they are left out; a closure that has finished stays in the feed for seven days, so we leave out any whose end time has passed. A closure is described as closing the carriageway when the service says so, and as a lane or slip closure otherwise. Roads come from the service’s own road number, so they match our road pages exactly, and a scheme naming several roads counts on each of them.

Roadworks in Scotland come from the Scottish Road Works Register’s open data: one extract a day, taken overnight, of the works, events and permissions recorded on Scotland’s roads. It is a snapshot, not a live feed, so works listed as under way may have finished and works recorded after the extract are not shown. Every page says which day’s extract it is using.

We count one set of works (the register calls it an activity) once. Works are under way when the register’s status is In Progress or Commenced, and planned otherwise; ’potential’ works, which are only an early notice that something may happen, and plant enquiries are left out. A closure means traffic management that closes a road, a lane, a slip road or the hard shoulder, or runs a contraflow.

Works belong to a council when the register names that council as the road authority. Transport Scotland’s operating companies and DBFO contracts look after the trunk roads, so their works count nationally and on the road’s own page but on no council page. Works are matched to a road only when the register’s street name starts with the road number, so road counts are lower than the true figure rather than higher.

Dates, descriptions and traffic management are recorded by whoever is doing the work, and the register does not guarantee the files are free of errors. Roadworks in England are held in DfT’s Street Manager, which sends events to a subscriber’s own server rather than publishing a file, and Wales publishes separately; neither is connected yet, so our roadworks figures cover Scotland only.

EV charging

Public charger counts come from DfT’s quarterly electric vehicle charging infrastructure statistics (tables EVCI0101, EVCI0102 and EVCI0201), which are based on Zapmap data and are official statistics in development. They count chargers operational at midnight on each date. Council figures start in January 2026 and region and nation figures in January 2025, as DfT publishes them; county figures are the sum of their councils. “Rapid” means 50kW and above. Chargers per 100,000 people use the ONS mid-2024 population, as DfT does.

Private plug-in cars per charger divides the privately kept battery electric, plug-in hybrid and range-extended cars licensed at the end of a quarter by the public chargers on the first day of the next (the end of March against 1 April). We use privately kept cars because company cars are registered at the company’s address, which would put whole leasing fleets into one council.

Local figures

Local vehicle counts come from DfT/DVLA tables VEH0105 (licensed vehicles by body type, fuel and keepership) and VEH0142 (plug-in vehicles), for every UK council, county, region and nation, quarterly from the end of 2009 (Northern Ireland from September 2014). Petrol, diesel and petrol hybrid counts come from VEH0105; battery electric and plug-in hybrid counts from VEH0142; “other fuels” is VEH0105’s other-fuel count less the plug-in cars. Trends use the end of March each year.

Vehicles are counted at the registered keeper’s postcode, so company, lease and rental cars appear where the company is registered, and totals move when a large fleet changes address. We therefore show privately kept cars alongside all cars, and the EV page ranks councils on privately kept cars. Cars per 1,000 people divide licensed cars by the ONS mid-year population for the same area (cars at the end of June for the yearly series).

Neighbourhood figures come from DfT tables VEH0125 and VEH0145, which give licensed and plug-in vehicles for each 2021 lower layer super output area (LSOA). In England and Wales an LSOA’s name gives its middle layer super output area (MSOA, about 7,000 residents) and its 2021 council, so we add LSOAs up to MSOAs and place them in today’s councils. We show the battery electric share of privately kept vehicles of all types. Areas with a suppressed small count are not ranked. Scottish data zones and Northern Irish super data zones cannot be placed in councils from their names, so they are left out.

Safety recalls

Recalls come from DVSA’s file of every recall since 1992, which has one row for each recall and model, to 14 September 2026. Manufacturers notify recalls to DVSA; DVSA says the details may be delayed or incomplete, so every page points to GOV.UK’s vehicle recall check, which looks up a specific car.

A recall number starts with its type: R for road vehicles (cars, vans, trucks and buses), RM for motorcycles, and others for components, special purpose vehicles and trailers. We use R/ recalls (7,495 since 1992) and match each row’s make and model to DVLA’s car models. Company words come off the make (“TOYOTA (GB) PLC” is Toyota), group recalls are tried against each of the group’s makes (“FIAT CHRYSLER AUTOMOBILES UK LTD” covers Fiat, Abarth, Alfa Romeo and Jeep), and a model name listing two models (“S-MAX/GALAXY”) counts for both. Model names then go through the same rules as MOT results: suffixes such as “PHEV” or “MCA” are dropped, a tiny stray DVLA entry does not beat a big one, and a name DVLA splits by engine maps only to a version that holds most of the cars. Because DVLA counts only cars under a model name, a longer name (“TRANSIT COURIER”) wins over the name it extends (“TRANSIT”). 94.7% of rows from car makes match a model. The rest include vans and trucks sold under car makes, names given only as a range (“MINI”) and models DVLA does not list yet.

A car recall is one that names at least one car model. Some models are mostly registered as vans (DVLA counts cars and vans under the same names): where more than half are vans, as with the Ford Transit and Mercedes-Benz Sprinter, the model’s page still lists its recalls, with a note, but the model is left out of the car counts and rankings. From 15% vans, pages say the recalls may cover vans too.

DVSA wrote most concerns in capitals. We put them in sentence case, keeping initialisms (ABS, ESP, ECU, SRS, VIN) and make and model names. DVSA’s export lost some apostrophes, leaving a replacement character: where it sits between a letter and an s or t (“passenger’s”, “doesn’t”) we put the apostrophe back; any other one is dropped, and each is listed for review. One recall has a launch date in the future; we use the same day in the year of its recall number, and never publish a future date. VINs are not published.

Vehicles affected is the number the manufacturer gave DVSA. For most recalls covering several models, DVSA repeats the recall’s total on every row, so a model gets its own figure only when the rows give one per model. Build dates are the manufacturer’s; overlapping periods for the same model are joined.

Model pages give the number of recalls and the latest in full. A model gets its own recall list when it has at least 3 recalls; the list can be indexed from 5 recalls launched in at least 2 different years. Model-year pages list the recalls whose build dates include any day of that year. The year of a model-year page is the year of first registration, not the year the car was built, so a car first registered early in a year may have been built the year before. We do not build a page for each recall: it would repeat one paragraph of DVSA’s text, which GOV.UK already shows.

Counts of recalls are not ratings. A make with more models, more versions and more cars on the road has more recalls, manufacturers split recalls differently, and one recall can cover a handful of cars or hundreds of thousands.

Driving tests

Pass rates come from DVSA tables DRT122A (car practical tests by centre, month and gender), DRT122C (first attempts) and DRT121F (the ten most common faults), for Great Britain in financial years from 2007/08. A centre is followed across years by its name, ignoring case, spacing and the “(London)” DVSA later added; any other rename starts a new record. Closed centres are left out. DVSA leaves out centres where only one examiner tested (their tests still count nationally) and withholds small counts. We rank pass rates only among centres with at least 1,000 tests in the year, and do not build a page for a centre with fewer than 100. DVSA’s tables give centre names only, so we do not place centres in council areas.

When we publish a page

Before we build a page for a model, road or place, we score the data behind it for size, history, number of measures, how different it is from related pages, and confidence. Pages that pass are published and indexed. Pages close to the line are published but kept out of search results. Pages below it are not built, and links point to the nearest page that covers them.

For example, a car model needs at least 1,000 records, 3 years of history and 5 distinct measures before its page can be indexed.

Scores and indices

We plan three scores: the UK Motoring Reliability Score, the Road Danger Index and the EV Infrastructure Score. None has been published yet. Each will appear only after its method is set out here in full, with inputs, weights and limits.