Data & method

The dataset

All economy figures come from the US EPA / DOE fuel economy database, downloaded in bulk from fueleconomy.gov — 49,995 vehicle records covering model years 1984 to 2027. Retrieved 2026-07-28.

A straight note about the licence

fueleconomy.gov publishes no explicit licence or terms of use for this dataset. It is a work of the US federal government — produced by the EPA and DOE and hosted by Oak Ridge National Laboratory — and is therefore treated here as public domain under 17 U.S.C. § 105. That is a reasoned position, not a licence grant, and it is stated plainly rather than dressed up as one. If the agency asks for anything to change, it changes.

How models are rolled up

The raw data has one row per trim, engine and drivetrain — six separate rows for a single year of Corolla. Publishing all 26,142 make/model/year combinations would produce thousands of near-identical pages, so records are rolled up to make and model, keeping the full year-by-year history on each page. Each year's figure is the average across all EPA-tested trims that year, and the trim count is shown so you can see how much averaging sits behind it.

Only models with five or more model years of data are published, which is what makes the drift-over-time analysis meaningful.

The fuel price

Cost per mile and annual fuel spend use a reference price of $3.20 per gallon over 15,000 miles a year. That number is an assumption, it is labelled as one on every page, and the calculator lets you replace it. The EPA's own published annual cost figure — which uses its own separate price assumption — is shown alongside in the year table, so you can see both.

Electric vehicles

Electric models are rated by the EPA in MPGe, the distance travelled on the energy equivalent of a gallon of petrol. It is a useful efficiency comparison and a terrible proxy for running cost, because charging is billed per kilowatt-hour. Dollar figures on EV pages are labelled as petrol-equivalent illustrations only.

What these figures cannot tell you

• They are laboratory test results. Real-world economy is routinely 10–20% worse, and worse again in cold weather, in traffic, towing, or with a heavy right foot.
• Test procedures have changed. EPA methodology was revised substantially in 2008 and adjusted since. Comparing a 1990 figure with a 2025 one is not quite like for like, which is worth remembering on the long drift tables.
• Fuel is not the cost of ownership. Depreciation usually dwarfs it, and insurance, tyres, servicing and repairs are all absent here.
• Trim averaging hides spread. A model year offering both a four-cylinder and a V8 gets one averaged number on the summary rows; the best and worst columns show how wide that spread is.

Estimates based on official test figures and average fuel prices. Real-world consumption varies with driving style, load, terrain and maintenance.

Corrections

If a figure disagrees with fueleconomy.gov, fueleconomy.gov is right. Send the page URL.