Nuclear encyclopedia

Chapters / Accidents and safety

Accidents and safety

Three Mile Island, Chernobyl, Fukushima, the smaller events, and what the safety record actually shows.

Status: first sections published with citations; INES register and sequences to come. Release v0.8.1, 20 November 2026. Bracketed [source-id] links point to the source register.

Three accidents define public understanding of nuclear power, and each taught a different lesson. Three Mile Island (1979) was a partial meltdown contained by the building around it: small release, enormous political damage. Chernobyl (1986) was a design with a positive void coefficient pushed outside its operating envelope during a test, with no containment: the worst nuclear accident in history. Fukushima Daiichi (2011) was a station blackout after an earthquake and tsunami that exceeded the design basis: a reminder that decay heat, not the chain reaction, is the persistent hazard.

Measured per unit of energy produced, nuclear power's death toll is low, comparable to wind and solar and far below coal, though estimates of Chernobyl's long-term toll span a wide range and are presented here with their sources and methods. The chapter also covers the INES scale, near-misses, and the safety reforms each accident forced.

Three Mile Island, 1979: contained, not harmless

On 28 March 1979 the Unit 2 reactor at Three Mile Island partially melted down after a stuck valve, misleading instrumentation and operator error let the core uncover. The containment held: the roughly two million people nearby received an average additional dose estimated at about 1 millirem, against a natural background of 100-125 millirem a year, and studies by the NRC, EPA, Pennsylvania and independent universities found no detectable health effects. The NRC calls it the most serious accident in US commercial nuclear history, and its real legacy: emergency planning, operator training, human-factors engineering and a complete reworking of oversight. [nrc-tmi]

Chernobyl, 1986: the worst case, measured

The explosion of RBMK reactor 4 on 26 April 1986 remains the only nuclear accident with immediate radiation deaths among the public-facing record: of 600 workers on site that morning, 134 suffered acute radiation sickness and 28 died within three months; 19 more of that cohort died by 2004 of causes not necessarily attributable to radiation. Among residents of Belarus, Russia and Ukraine, more than 6,000 thyroid cancers were diagnosed by 2005 in people exposed as children, most of them attributable to radioiodine in milk; UNSCEAR finds no evidence of a major public-health impact beyond this, two decades on. [unscear-chernobyl]

For long-term mortality the sources must be quoted with their methods, because they differ. The WHO-led Chernobyl Forum projected up to about 4,000 eventual radiation-induced deaths among the roughly 600,000 most exposed people (emergency workers, evacuees and residents of the most contaminated areas), an increase of about 3 per cent over the spontaneous cancer toll in that cohort, and stressed it would be difficult to observe statistically. Higher published estimates exist; they use different dose models and are noted on the gaps page rather than reconciled here. [who-chernobyl-2005]

Fukushima Daiichi, 2011: radiation versus evacuation

The 11 March 2011 earthquake and tsunami cut all power to the Fukushima Daiichi site, and three reactors melted down over the following days. UNSCEAR's 2020/2021 assessment, ten years on: no adverse health effects among residents have been documented that are directly attributable to radiation exposure, and future radiation-related effects are unlikely to be discernible; the apparent rise in childhood thyroid cancers is attributed to ultra-sensitive screening, and no credible evidence links radiation to birth outcomes. Worker doses were mostly low; a small fraction exceeded 100 mSv in the first year. [unscear-fukushima-2020]

The accident's death toll came from the response, not the radiation: the Japanese government counted 2,314 disaster-related deaths linked to Fukushima as of September 2020, overwhelmingly among elderly and hospitalised evacuees, a figure used by Our World in Data in its energy-death accounting. That trade-off, radiation risk against evacuation risk, is now central to emergency-planning doctrine. [owid-deaths]

Deaths per terawatt-hour: the comparative record

Our World in Data's accounting (accidents plus air pollution, 1965-2021) puts nuclear at 0.03 deaths per TWh: built from 433 attributed Chernobyl deaths (its own upper-bound calculation) and 2,314 Fukushima disaster-related deaths, divided by 96,876 TWh of cumulative nuclear generation. The same table: coal 24.62, oil 18.43, biomass 4.63, gas 2.82, hydropower 1.30 (dominated by the 1975 Banqiao dam failure, about 171,000 deaths), wind 0.035, solar 0.019. Fossil figures come from Markandya & Wilkinson (2007, The Lancet) for European plants; the fossil numbers are, if anything, understated for the world fleet. [owid-deaths]

Still to come in this chapter

  1. The INES register: every event rated 4 or above, including Windscale 1957, Kyshtym 1957 and Tokaimura 1999
  2. The three accident sequences in full, hour by hour, from INSAG-7 and the IAEA Fukushima report
  3. Post-Fukushima stress tests, filtered vents, and the safety case made for Gen III+ designs
  4. Occupational dose trends: WANO and UNSCEAR worker-exposure data

Claims in the final text will carry citations to the source register; anything unverifiable will land on the gaps page instead.