Is nuclear energy safe?

"Is nuclear energy safe?" has no one-word answer, but it does have an honest way to approach it with numbers: how many deaths each energy source causes per unit of electricity generated, counting both accidents and the air pollution that each technology routinely releases. That comparison avoids two symmetrical mistakes: judging nuclear only by its most visible accidents, without weighing that other sources kill quietly and continuously, or looking only at the average and erasing the memory of Chernobyl and Fukushima. The table below brings together the two peer-reviewed academic sources used today for this comparison across energy sources.

Markandya and Wilkinson (The Lancet, 2007) calculated deaths from air pollution and accidents for coal, oil, gas, biomass and nuclear, using European electricity-production data; between 88% and 99% of those deaths are chronic effects rather than one-off accidents, and the nuclear figure comes from a theoretical model — linear, no threshold — which assumes that any dose of radiation, however small, carries a proportional risk, plus occupational deaths from mining and milling; the study predates Fukushima. Sovacool and colleagues (Journal of Cleaner Production, 2016) instead built a database of energy accidents from 1950 to 2014 limited to low-carbon sources — it does not cover coal, oil or gas — that does include hydropower, wind, solar, biomass and nuclear, but counts only deaths from recorded accidents, without the theoretical projection of future cancers: for nuclear, Chernobyl, Fukushima and Kyshtym; for hydropower, the calculation period (1990-2013) leaves out the 1975 collapse of China's Banqiao dam, the deadliest hydropower accident on record, whose inclusion would multiply the rate by orders of magnitude. That is why the table marks each row with the methodology it uses: air pollution plus accidents for coal, oil, gas, biomass and one of the two nuclear figures; recorded accidents only for hydropower, wind, solar and the other nuclear figure. The gap between the two nuclear figures is a factor of seven: one counts documented accidents, the other adds a long-term estimate on top. The same pattern shows up for biomass, though the table publishes only the air-pollution-inclusive figure: 4.63 deaths per TWh, against 0.0164 counting only recorded accidents. Even at the higher nuclear estimate, it sits far below fossil fuels: the gap is explained mostly by air pollution, which thermal power plants produce every day, against a nuclear risk concentrated in rare catastrophic events.

Chernobyl, on 26 April 1986, is a level 7 on the INES scale, the maximum. According to UNSCEAR, the accident caused the deaths, within a few weeks, of 30 workers; of the 134 who suffered acute radiation sickness, 28 died within the first three months. The Chernobyl Forum (IAEA, WHO and other UN agencies, 2005) projected about 4,000 eventual deaths among the 600,000 most exposed people — 200,000 emergency and recovery workers from 1986-87, with 2,200 deaths expected in that group alone; 116,000 evacuees; and 270,000 residents of the most contaminated areas — plus about 4,000 thyroid cancer cases in children and adolescents, with at least nine deaths and a survival rate near 99%. UNSCEAR itself is more cautious: beyond the rise in thyroid cancer — more than 6,000 cases documented by 2005 — and an emerging signal of leukemia among the most exposed workers, it states there is no demonstrated increase in solid cancers or leukemia in the broader exposed population, and that such an increase would be very difficult to detect in epidemiological studies given the doses involved. The gap between the Chernobyl Forum's 4,000 deaths and UNSCEAR's caution is not a disagreement over the facts: it is the difference between a lifetime projection using the linear no-threshold model applied forward onto an exposed cohort, and a report of what epidemiological studies have managed to confirm statistically so far.

Fukushima Daiichi, on 11 March 2011, is also a level 7, though with a far smaller release than Chernobyl's, as the INES scale entry explains. There were no acute radiation deaths among workers or the public. The UNSCEAR 2020 Report concludes that, since the UNSCEAR 2013 Report, "no adverse health effects among Fukushima residents have been documented that could be directly attributed to radiation exposure from the accident"; the increase in thyroid cancers found in child screening is attributed to ultrasensitive screening revealing thyroid abnormalities that had not previously been detected, not to radiation. The documented toll of the accident is mostly indirect: Japan's Reconstruction Agency counted 2,348 "disaster-related deaths" in Fukushima prefecture through the end of 2024, deaths recognized as caused by the disaster through the aggravation of an injury or the physical burden of evacuation life; that category lumps together the effects of the earthquake and tsunami with those of the nuclear evacuation, and the source does not separate them.

Three Mile Island, on 28 March 1979, is a level 5: a large part of the core melted, but containment held. Decades of studies by the NRC and independent universities found no deaths or measurable health effects attributable to the radioactivity released. The roughly two million people in the surrounding area received an average dose of only about 0.01 mSv (1 millirem) above background, against about 0.06 mSv (6 millirem) from a single chest X-ray and 1 to 1.25 mSv (100 to 125 millirem) a year of natural background radiation in the area.

What these numbers show: across every energy source compared, generating electricity costs lives, and it almost always does so quietly and chronically rather than spectacularly; the three severe nuclear accidents documented since commercial generation began killed very few people directly compared with what the table attributes to air pollution from fossil fuels.

What these numbers do not show: whether that comparison covers everything needed to decide on a particular energy source, or whether it offsets other risks — waste, proliferation, technological dependence — that do not fit into a table of deaths per terawatt-hour.