How Dangerous Is Nuclear Power?

Counting accidents makes a source that produces little electricity look safe because it is small, so the standard measure is deaths per terawatt-hour of electricity produced. A terawatt-hour is about the electricity 150,000 people in the European Union use in a year.

The fossil fuel figures, from Markandya and Wilkinson (2007, The Lancet), are mostly not accidents but deaths from air pollution. The pollution is chiefly small particles, many formed from the sulphur dioxide and nitrogen oxides that burning releases. Of coal’s 24.62 deaths per terawatt-hour, accidents account for 0.12. Markandya and Wilkinson assessed European plants with modern pollution controls. So for much of the world’s coal fleet, the gap between coal and nuclear is if anything understated.

Nuclear’s 0.03 is the opposite: Our World in Data, produced with researchers at the University of Oxford, built it entirely from deaths at Chernobyl and Fukushima. It counted 433 at Chernobyl, the upper end of its own calculation, and 2,314 at Fukushima, the toll as of September 2020. All but one of the Fukushima deaths were among evacuees.

Two other methods give other figures, and all three are more than 30 times below gas. Markandya and Wilkinson put nuclear at 0.074 in Europe, mostly estimated cancer deaths plus deaths in mining, milling and generation. Sovacool et al.’s accident database (2016, Journal of Cleaner Production) gives about 0.01, as Our World in Data tabulated it in 2020.

Hydropower’s 1.30 comes almost entirely from the failure of the Banqiao dam in China in 1975. Our World in Data counts that failure as about 171,000 deaths, most of them later deaths from epidemics and famine rather than drowning. Without Banqiao, hydropower falls to about 0.04.

Chernobyl’s reactor was an RBMK-1000, a Soviet design whose Russian initials stand for high-power channel reactor. It had flaws most Western reactors lack: its nuclear reaction sped up when its cooling water turned to steam, and it had no Western-style containment. Early on 26 April 1986, during a turbine test, too many control rods had been removed and automatic shutdown systems disabled. An attempt to shut the reactor down then set off a power surge, and steam explosions destroyed it.

The Chernobyl Forum was eight United Nations agencies with the governments of Belarus, Russia and Ukraine. In 2005 it put the deaths directly attributed to radiation at fewer than 50, almost all among highly exposed rescue workers. Two workers were killed on the night of the accident, from injuries unrelated to radiation. Of 134 emergency workers and staff with acute radiation syndrome, 28 died in 1986. Another 19 died by 2006, usually of causes not associated with radiation, the United Nations Scientific Committee on the Effects of Atomic Radiation (UNSCEAR) reported.

Chernobyl’s counted toll also includes thyroid cancer in people exposed as children or adolescents, many through milk contaminated with radioactive iodine. Between 1991 and 2015, almost 20,000 thyroid cancers were registered in Belarus, Ukraine and the four most contaminated Russian regions among people who were under 18 in 1986. UNSCEAR estimated in 2018 that radiation caused about a quarter of the cases among those not evacuated, with an uncertainty range of at least 7 to 50 per cent. By 2005, 15 had proved fatal. Judging from experience in Belarus, the Chernobyl Forum put survival at almost 99 per cent.

The Chernobyl Forum projected that up to 4,000 people could eventually die of radiation-induced cancer among the roughly 600,000 most exposed. A WHO report the following year added up to 5,000 more, among five million people in less contaminated areas. That gave up to 9,000 in all. The International Agency for Research on Cancer projected about 16,000 cancer deaths across Europe by 2065, in a peer-reviewed study by Cardis and colleagues (2006, International Journal of Cancer). The study’s 95 per cent range runs from 6,700 to 38,000.

The two largest figures on the chart come from a report commissioned by a German Green member of the European Parliament and from Greenpeace. The TORCH report, by Ian Fairlie and David Sumner, gave 30,000 to 60,000. Greenpeace gave about 93,000, from another researcher’s estimate built on Belarus’s official health statistics.

The projections differ in how many lightly exposed people they include and in the risk per dose they assume. Radiation doses are measured in sieverts, a unit that reflects the damage radiation does to living tissue. A collective dose adds up the doses of everyone exposed, in person-sieverts.

TORCH’s method multiplies a collective dose by a risk of about 5 to 10 fatal cancers per 100 person-sieverts. That risk is taken from people such as the atomic bomb survivors, who received much higher doses. TORCH starts from 600,000 person-sieverts. At the lower risk that gives 30,000 deaths, and at the higher 60,000.

Higher risks per dose give higher tolls. In a 1990 book, the Berkeley chemist and radiation researcher John Gofman projected about 475,000 fatal cancers, assuming a risk of about 37 per 100 person-sieverts. Greenpeace’s report lists others higher still. A toll of 475,000 would put nuclear’s rate at about 5 deaths per terawatt-hour, above gas.

In the general population UNSCEAR has found no persuasive evidence of any radiation effect beyond thyroid cancer. There are indications, though, of more leukaemia and cataracts among recovery workers who received higher doses. An effect can exist without being detected: the linear no-threshold model predicts an excess too small to see against the normal variation in cancer rates.

Greenpeace’s report also cites an estimate of 200,000 extra deaths from all causes in Belarus, Russia and Ukraine between 1990 and 2004. It comes from a 2006 Russian study of health statistics. In a 2009 book, Alexey Yablokov of the Russian Academy of Sciences describes the study as comparing death rates in contaminated provinces with those in less contaminated ones. Applying its rate, and assumed fractions of it, to people across Europe and the world, he reached 985,000 extra deaths from April 1986 to the end of 2004.

A toll of 200,000 would put nuclear’s rate at about 2 deaths per terawatt-hour, still below gas. One of 985,000 would put it at about 10, above gas but still well below coal. UNSCEAR, however, regards comparisons between areas as weaker evidence than studies of individual people. It also notes that death rates rose in most of the former Soviet Union in recent decades, which studies of the accident must take into account.

Download this conclusion: Chernobyl’s toll

This card takes apart the claim that Chernobyl killed tens or hundreds of thousands. By 2005 fewer than 50 deaths had been attributed to radiation, plus 15 from thyroid cancer. The large figures it shows, from 4,000 to about 93,000, are projections of eventual cancer deaths. Anyone may download it and post it.

Download the card (PNG, 1,200 × 630)

On 11 March 2011 a magnitude 9.0 earthquake and the tsunami that followed killed 15,901 people in eastern Japan, by the National Police Agency’s count in March 2026. Another 2,519 are still missing. The tsunami disabled the power supply and cooling of three reactors at the Fukushima Daiichi plant. All three cores largely melted within three days.

No member of the public has died of radiation from Fukushima, and no case of acute radiation sickness occurred. One death has been officially recognised as caused by radiation: in September 2018 Japan’s labour ministry granted workers’ compensation for a man in his fifties. He had measured radiation at the plant after the accident and later died of lung cancer. The ministry put his exposure at about 195 millisieverts, thousandths of a sievert, from radiation work at several plants, about 74 of them after the accident. It was a compensation decision, not a clinical finding.

The United Nations Scientific Committee on the Effects of Atomic Radiation (UNSCEAR) published its 2020 report in 2021. It concluded that no adverse health effects among Fukushima residents have been documented that could be directly attributed to radiation. It also judged that future health effects from radiation are unlikely to be discernible. Fukushima prefecture’s health survey estimated external doses in the first four months for 421,394 residents. Of these, 62 per cent received less than 1 millisievert. UNSCEAR puts the worldwide average dose from natural sources at 3.0 millisieverts a year.

Japan’s Reconstruction Agency counted 2,350 disaster-related deaths in Fukushima prefecture by the end of 2025. These are deaths from the strain of the disaster and of evacuation, not from the earthquake, the tsunami or radiation. Not all can be separated from the earthquake and tsunami. But the agency’s 2012 review found more such deaths in Fukushima prefecture than in any other, and put that down largely to the evacuation.

That review examined 734 Fukushima deaths. About nine in ten were of people aged 70 or over. Seven in ten came within three months. The main causes were exhaustion from shelter life and from moving, and delayed treatment when hospitals stopped working.

Critics question whether so large an evacuation was needed, since it contributed to thousands of early deaths and radiation has been linked to one. UNSCEAR, however, estimated that the evacuation cut first-year doses in the evacuated areas by up to a factor of ten.

By June 2021, ultrasound screening had found 263 malignant or suspected thyroid cancers, far more than expected. It covered those aged 18 or under at the time, about 368,000 people, and later those born in the following year.

UNSCEAR attributes the excess to the screening, which finds many cancers that may never have caused symptoms. It points out that, unlike after Chernobyl, no excess appeared in children exposed before the age of five. It also notes that the cancers appeared within one to three years, rather than from four to five years as after Chernobyl and in other radiation studies.

Some researchers, among them Tsuda et al. (2016, Epidemiology), argue that the excess is unlikely to be explained by screening.

About 22,400 people from Fukushima prefecture were still evacuated in August 2026, down from about 160,000 in May 2012. By July 2020 only about 30 per cent of former residents had returned to the twelve evacuated municipalities, in Japan’s Board of Audit’s figures.

Download this conclusion: Fukushima’s deaths

This card takes apart two opposite claims, that nobody died at Fukushima and that it proves nuclear power is unsafe. One worker’s death is officially recognised as caused by radiation. Separately, 2,350 disaster-related deaths, mostly linked to the evacuation, were certified in Fukushima prefecture by the end of 2025. Anyone may download it and post it.

Download the card (PNG, 1,200 × 630)

At Three Mile Island in 1979 at least 45 per cent of the core of Unit 2 melted, and the containment building held. The US Nuclear Regulatory Commission (NRC) reports an estimated average dose of about 0.01 millisieverts to the roughly two million people around the plant. That compares with a natural background there of about 1 to 1.25 millisieverts a year.

No radiation death from the accident has been confirmed. A University of Pittsburgh follow-up of 32,135 people living within five miles found no consistent evidence that the release affected cancer deaths to 1998. Some cancers were raised in the higher exposure groups, though. And a 1997 reanalysis of cancer incidence near the plant, by Wing and colleagues, found lung cancer and leukaemia rising with estimated dose.

Three other fatal accidents are sometimes set against nuclear power, though none happened at a civilian power plant.

The 1999 accident at Tokaimura in Japan, which killed two workers, did not happen at a civilian power plant either: it was at a plant preparing fuel for experimental reactors. In the history of civilian nuclear electricity, Chernobyl is the only accident with a substantial confirmed radiation death toll.

Reactors worldwide had discharged about 448,000 tonnes of spent fuel by June 2026, by the International Atomic Energy Agency’s count as the American Nuclear Society reported it. The United States holds about 90,000 tonnes, at more than 70 sites. Those sites are in 35 states, and the Department of Energy says all of that fuel would fit on a single football field less than 10 yards deep. A small volume of intensely radioactive material is a different problem from a large volume of a mild one, not automatically an easier one.

Spent fuel is held first in cooling pools and then in dry casks, sealed steel cylinders inside concrete or steel shielding. The first American dry casks were loaded in 1986. Since then, the US Nuclear Regulatory Commission (NRC) states, dry storage has released no radiation that affected the public or contaminated the environment. That record is short against the timescales involved, though, and the NRC has studied canister ageing.

Finland is closest to permanent disposal. Posiva’s Onkalo repository at Olkiluoto places spent fuel in iron and copper canisters packed in bentonite clay about 430 metres down in the bedrock. In August 2026 Finland’s Radiation and Nuclear Safety Authority found that the repository meets the safety requirements for an operating licence. Posiva aims to be ready to begin disposal by the end of 2026. Onkalo would be the first deep geological repository for spent fuel in operation anywhere.

In the United States, Congress designated Yucca Mountain for a repository in 1987, and the Department of Energy moved to withdraw its licence application in 2010. The NRC finished its safety evaluation in January 2015. It found that the application met its requirements, which run to a million years, apart from land and water rights the department had not obtained. No waste was ever emplaced. In July 2026 the department turned instead to five states that had offered to explore hosting nuclear waste, the Las Vegas Sun reported.

Short-lived fission products such as caesium-137 dominate the hazard for about the first hundred years. After that, long-lived actinides, mainly plutonium and americium, control a long tail. Spent fuel disposed of directly falls to the radiotoxicity of the uranium ore it came from only after more than 100,000 years, by the Nuclear Energy Agency’s reckoning. The 10,000-year figure often quoted was a regulatory compliance period, not the point at which the material becomes harmless.

Cost and time explain why so little nuclear power is being built in Western countries, even if it is as safe as the mortality data suggest. The four Western projects in the chart ran years late and cost roughly two to seven times their first budgets. Westinghouse, the reactor supplier for Vogtle in the United States, filed for bankruptcy protection in 2017.

The 2026 analysis by Lazard, a financial advisory and asset management firm, compares the lifetime cost of electricity from new plants, without subsidies. It puts new nuclear at 175 to 255 dollars per megawatt-hour, against 37 to 99 for onshore wind. Large solar farms come to 40 to 98, and combined-cycle gas to 51 to 129. Running an existing, fully depreciated American nuclear plant costs 26 to 36 dollars, which is why keeping old reactors open is a very different proposition from building new ones.

Because nuclear’s cost is largely paid upfront, it depends heavily on the cost of borrowing. The Nuclear Energy Agency and the International Energy Agency published a costing study in 2020. It put a new French reactor at 45 dollars per megawatt-hour at a 3 per cent discount rate. At 10 per cent, the same reactor came to 97.

Against that, the wind and solar figures leave out the cost of covering periods without wind or sun. Lazard estimates that cost under the rules of six American grid operators. It brings wind and solar, with or without batteries, to 87 to 167 dollars, still below new nuclear. But its method does not price a supply that runs around the clock.

The 2020 costing study also found that nuclear construction is not inherently slow or expensive. It put the overnight cost of a new plant, the cost before any interest, at about 2,157 dollars per kilowatt in South Korea and 2,500 in China. In the United States it was 4,250. Whether Western countries can build at Korean costs is not settled.

Comparing 349 reactors in seven countries, Lovering, Yip and Nordhaus (2016, Energy Policy) suggest that nuclear technology has no inherent trend of rising costs. France built most of its fleet quickly in the 1970s and 1980s. Grubler (2010, Energy Policy) found that centralised decision-making, standardisation and stable regulation were key to that speed, though even there real construction costs rose substantially.

Critics such as Amory Lovins add that nuclear power is slow. Each dollar spent on a new reactor, he argues, buys about two to ten times less carbon saving than cheaper alternatives, and buys it twenty to forty times more slowly. A 2019 report by the International Energy Agency, though, found that in advanced economies, replacing reactors with renewables would cost more than extending the reactors’ lives.

Civilian programmes have also fed weapons programmes. India’s 1974 test probably used plutonium from a research reactor built with Canada, whose heavy water the United States had supplied for peaceful use. North Korea built its plutonium weapons around its Yongbyon complex. A network run by Abdul Qadeer Khan, often called the father of Pakistan’s nuclear weapons programme, supplied uranium-enrichment technology to Iran, Libya and North Korea.

The International Atomic Energy Agency’s safeguards check, through inspections and nuclear accounting, that peaceful nuclear material is not diverted to weapons. They were strengthened after Iraq’s secret programme was discovered. The Additional Protocol of 1997, which countries may add to their safeguards agreements with the agency, aims to detect cheating much earlier. But detection is not prevention. This cost is not measured in deaths per terawatt-hour.

Replacing nuclear generation with coal raised air pollution deaths in Germany by about 330 to 800 a year from 2012 to 2019, by a peer-reviewed estimate. The range depends on the method.

Germany accelerated its nuclear exit after Fukushima, closing eight reactors in 2011 and the last three on 15 April 2023. Coal’s share of German electricity rose from 24.8 per cent in 2020 to 33.2 per cent in 2022. The federal statistics office says coal helped make up for sharp falls in power from nuclear and gas, whose price rose sharply after Russia invaded Ukraine.

Jarvis, Deschenes and Jha’s 2019 working paper put the cost of the phase-out at about 12 billion dollars a year. More than 70 per cent of that came from about 1,100 extra deaths a year caused by air pollution. Their peer-reviewed version appeared in the Journal of the European Economic Association in 2022. It lowered the cost to 3 to 8 billion euros a year. It also lowered the extra deaths to 330 to 800 a year, with 800 as the authors’ preferred figure.

A separate 2025 study estimated roughly 170 extra deaths a year before the age of 75 from non-infectious respiratory diseases alone. Another, in the journal PLOS One, found that the phase-out’s extra yearly carbon dioxide emissions had disappeared by 2023 as renewables grew.

Far less dangerous than fossil fuels. Per unit of electricity, nuclear power is among the safest sources, level with wind and solar in Our World in Data’s figures and hundreds of times safer than coal. The fossil fuel deaths come from the air pollution of normal operation, while nuclear’s come from two accidents. Even Greenpeace’s projection of Chernobyl’s toll would leave nuclear below gas; John Gofman’s, which assumes several times the risk per dose that other projections use, would not. And at Fukushima all but one of the deaths came from the disaster and the evacuation, not from radiation.

What counts against nuclear power is mainly its cost and building time, with permanent waste disposal and the spread of weapons still unsettled.

Download the conclusion

The verdict above as one image, with the three numbers that frame it. They are deaths per terawatt-hour of electricity in Our World in Data’s figures: 24.62 for coal, 2.82 for gas and 0.03 for nuclear. Anyone may download it and post it.

Download the card (PNG, 1,200 × 630)