EPA Fuel Economy Dataset (1984-2026)

Since model year 1984, every car, truck and van sold in the United States has been tested and rated for fuel economy, and the results sit in one continuous table. The EPA Fuel Economy Dataset (1984-2026) is that table: one row per vehicle configuration - a specific year, make, model and engine/transmission combination - carrying laboratory-measured efficiency, emissions and powertrain detail across 43 consecutive model years.

Figures originate at the EPA National Vehicle and Fuel Emissions Laboratory in Ann Arbor, Michigan, supplemented by manufacturer-run tests performed under EPA oversight; Oak Ridge National Laboratory administers the program for the US Department of Energy and EPA. For automobile manufacturers that makes it the longest continuous efficiency-and-emissions panel available for the US market - the yardstick every rival lineup gets measured against, model year after model year.

Get a sample of this dataset and the first rows arrive with your model-year filter already applied.

Rows ship flat - one configuration per line, no joins required. Illustrative rows in the delivered column order:

manufacturer: Toyota        carline name: Camry             class: Midsize Cars
year: 2023      displ: 2.5      cyl: 4      trans: Automatic 8-spd       drv: F
cty: 28    hwy: 39    cmb: 32        ucty: 33    uhwy: 47    ucmb: 38
fl: R      fcost: 1550     eng dscr: 2.5L L4 DOHC 16V     trans dscr: 8-speed automatic

manufacturer: Ford         carline name: F-150 Pickup 2WD  class: Standard Pickup 2WD
year: 2023      displ: 3.5      cyl: 6      trans: Automatic 10-spd      drv: R
cty: 18    hwy: 24    cmb: 20        ucty: 21    uhwy: 28    ucmb: 23
fl: P      fcost: 2900     eng dscr: 3.5L V6 Turbo        trans dscr: 10-speed automatic

Values above illustrate the shape of a row rather than quote a certified extract; your sample pulls live records. Electric configurations swap the displacement-driven MPG columns for MPGe equivalents and add charge-time detail, while diesel, ethanol and compressed-natural-gas variants ride through the same schema via the fuel-type code.

Twenty-three documented fields, verified against the publisher's own file-layout and web-services documentation. The dictionary below is the complete core schema for every configuration row; nothing is inferred.

Geography - the United States market exclusively: every row reflects EPA ratings of US-market vehicle configurations, and historical files additionally distinguish 49-state and California fuel formulations.

Temporal - model years 1984 through 2026 combined in a single table, with per-model-year packaging reaching back to 1978 and already carrying a preliminary 2027. Estimates for 1984-2007 and some 2011-2016 vehicles have been revised so the years read comparably side by side.

Granularity - one row per vehicle configuration: a year, make, model and engine/transmission option. A nameplate offering three engines and two transmissions occupies six rows, which is what lets lineup-level analysis sum cleanly.

API, files, or your warehouse. Daily, weekly, or hourly.

• Lineup benchmarking - plot each rival's city/highway/combined MPG by segment and model year, and quantify exactly where a powertrain strategy leads or trails.
• CO2 trajectories and compliance narratives - tailpipe-emissions series per manufacturer turn regulatory posture into a chart any board can read.
• Fleet and ownership-cost modeling - annual fuel cost rides on every configuration row, so total-cost-of-ownership calculators get tested inputs instead of assumptions.
• Machine-learning training corpora - four decades of tabular rows with a stable schema make this a canonical regression and classification benchmark for efficiency prediction.
• Residual-value and demand models - efficiency and fuel-cost fields serve as covariates explaining why two near-identical nameplates depreciate differently.
• Reporting and citation - the standard reference point for US fuel-economy claims in journalism, academic work and analyst notes.

Data scientists and ML engineers get a decades-deep, schema-stable tabular corpus that trains without augmentation. Competitive-intelligence and product teams get a per-segment, per-year read on every rival's engineering positioning. Market researchers and consultants get an authoritative forty-plus-year panel for category studies. Journalists, academics and students get the citation standard for US fuel-economy and emissions reporting. Investors and quant researchers get long-run efficiency trends by maker as evidence on technology risk. Developers building data products get flat files shaped to load into anything.

Four things worth knowing upfront. First, annual releases revise history: estimates for 1984-2007 and some 2011-2016 vehicles were recalculated for cross-year comparability, and revised estimates are flagged for several manufacturers including Nissan, Volkswagen Group, Hyundai/Kia and Ford - so re-pull rather than cache if your series spans those gaps.

Second, adjusted and unadjusted MPG are different series. The cty/hwy/cmb columns carry the window-sticker-style figures consumers recognize; ucty/uhwy/ucmb carry raw dynamometer results before adjustment. Keep them separate in any pipeline.

Third, two columns exist only for 2000-and-later model years - valves per cylinder and vehicle-class code - and interior-volume fields apply to body styles that have them. Treat pre-2008 rows as slightly thinner.

Fourth, the publisher publishes no row count alongside the combined table, so tens of thousands of configurations stays an estimate anchored on per-year packaging. Scope a sample to the years you care about and count what arrives.