Coal is by far the most dangerous power source per unit of electricity; nuclear is the safest, with natural gas in between

Summary

Coal is the most dangerous of the three per unit of electricity produced, natural gas is in the middle, and nuclear is the least dangerous by a wide margin. Using the standard 'deaths per terawatt-hour' metric — which combines accident deaths across the whole supply chain (mining, transport, construction, plant operation) with premature deaths from air pollution exposure — coal causes on the order of 25 deaths per TWh, natural gas about 2.8-3 deaths per TWh, and nuclear about 0.03 deaths per TWh (roughly 1 death every 33 years for a town consuming 1 TWh/year). Nearly all of the harm from coal and gas comes from air pollution (fine particulates, SO2, NOx) rather than accidents; nearly all of nuclear's tiny death toll comes from the two worst accidents in its history, Chernobyl (1986) and Fukushima (2011).

So ranked from most to least dangerous per unit of power produced: coal > natural gas >> nuclear.

The numbers and where they come from

The most widely cited peer-reviewed source for fossil-fuel death rates is Markandya, A. & Wilkinson, P. (2007), 'Electricity generation and health,' The Lancet, 370(9591): 979-990, which estimated life-cycle mortality (accidents plus air pollution) per TWh for European power plants:

• Coal: ~24.6 deaths per TWh
• Oil: ~18.4 deaths per TWh
• Natural gas: ~2.8 deaths per TWh
• Biomass: ~4.6 deaths per TWh

Nuclear's death rate is calculated differently, since there is no ongoing air-pollution toll: Our World in Data divides the total deaths attributed to the Chernobyl (433, per UNSCEAR) and Fukushima (2,314, per the Government of Japan's compensation/evacuation-related mortality figures) accidents by cumulative global nuclear electricity generation from 1965-2021 (96,876 TWh), giving roughly 0.03 deaths per TWh — about 800 times lower than coal and about 90 times lower than gas. Renewable and hydro accident-death data come from Sovacool et al. (2016), a database of energy-related accidents; hydropower's rate (1.3 deaths/TWh) is dominated almost entirely by a single catastrophic event, the 1975 Banqiao Dam failure in China (~171,000 deaths), and falls to ~0.04 deaths/TWh — comparable to nuclear — if that one event is excluded.

Why coal and gas differ so much despite both being fossil fuels

Coal combustion releases far more particulate matter, sulfur dioxide, and other pollutants per unit of energy than natural gas combustion, which burns cleaner and produces roughly half the CO2 and a small fraction of the particulate/SO2 pollution of coal per unit of electricity. That gap in air-pollution intensity is what drives coal's roughly 9x higher death rate compared with gas — it is overwhelmingly an air-pollution story, not an accident-rate story (coal mining accidents, while real, are a small share of the total compared to downstream population-level air-pollution mortality from plant emissions).

Important caveats on how conservative these numbers are

Our World in Data and the underlying literature flag that the Markandya & Wilkinson (2007) fossil-fuel figures are likely substantial underestimates of the true global average, for three reasons: (1) they are based on European plants with unusually strict pollution controls; (2) they don't reflect that plants in many low- and middle-income countries are sited much closer to dense population centers, increasing exposure; (3) they predate more recent epidemiological research (e.g. Lelieveld et al. 2018, which estimates roughly 3.6 million deaths per year globally from fossil-fuel air pollution) showing air pollution's health impact is more severe than 2007-era models assumed. Cross-checking the 2007 death rates against 2021 global electricity output implies about 280,000 deaths/year from fossil-fuel electricity — versus an estimated 1.1-2.55 million/year implied by more recent research, a 4-9x gap. Applying that correction would put the true global coal death rate in a range of roughly 93-224 deaths per TWh rather than ~25 — meaning the coal-vs-gas-vs-nuclear ranking almost certainly still holds (coal worst, gas better, nuclear/renewables far better), but the absolute gap between coal and gas, and between fossil fuels and nuclear, is probably even larger than the commonly cited chart shows.

Nuclear's own historical death toll also carries real uncertainty: some estimates of Chernobyl's ultimate cancer death toll (including long-tail radiation-linked cancers) run higher than the acute/near-term count used here, and Fukushima's count reflects deaths from evacuation-related stress and disruption rather than direct radiation exposure (essentially zero people are attributed to have died from Fukushima radiation itself). Both nuclear accidents remain vanishingly small next to the continuous, ongoing toll of fossil-fuel air pollution.

Data and methodology notes for this report

This is fundamentally a literature/epidemiology question rather than one this session's AskAmerica warehouse (search_catalog across the energy, health, and environment schemas) carries a direct 'deaths attributable to this power plant's emissions' table for — the warehouse holds EIA generation and EPA air-quality monitoring data (energy.eia_electricity_generation, environment.air_quality_annual, environment.egrid_emission_rates) but not a mortality-attribution model linking pollutant exposure to cause-specific deaths by fuel type at the plant level, which is exactly the kind of epidemiological modeling Markandya & Wilkinson (2007) and Lelieveld et al. (2018) perform. No AskAmerica query() calls were made for this report's headline figures; all numbers above were read directly from Our World in Data's published methodology and its cited primary sources (Markandya & Wilkinson 2007; Sovacool et al. 2016; Lelieveld et al. 2018; UNSCEAR).