Safety record, accidents and radiation doses

Nuclear safety arguments usually run on two tracks that never meet. One side points to the death rate per unit of electricity, which is very low. The other points to the consequences of a bad accident: whole towns emptied, land fenced off for decades. Both are true. An honest argument keeps both in view and gets the numbers right, which is harder than it sounds, because the numbers for nuclear accidents are some of the most mangled in public debate.

Three Mile Island, March 28, 1979

At Three Mile Island Unit 2 near Middletown, Pennsylvania, a pump failure in the non-nuclear part of the plant tripped the turbine and the reactor. A relief valve then stuck open, but a control room light told operators it had closed. Cooling water drained out unnoticed. Misreading their instruments, operators cut back emergency cooling. The core uncovered and about half of it melted.

The containment building held. The NRC says the roughly 2 million people nearby received an average of about 1 millirem above background, and the highest dose at the site boundary was under 100 millirem. Studies by Columbia University, the University of Pittsburgh, state and federal agencies found no adverse health effects that could be linked to the accident.

The consequences were large anyway. The accident shattered public trust, coincided with the release of a movie about a reactor cover-up, and changed U.S. regulation: resident NRC inspectors at every plant, rewritten operator training, a 24-hour NRC operations center, and the industry's own oversight group, the Institute of Nuclear Power Operations. Many reactor orders placed in the 1970s were cancelled in the years after, though rising interest rates and falling demand forecasts had already been killing projects.

Chernobyl, April 26, 1986

Chernobyl Unit 4 in Soviet Ukraine was an RBMK, a graphite-moderated, water-cooled design unique to the Soviet Union. During a poorly run safety test at low power, with safety systems disabled, power surged. The RBMK's positive void coefficient meant that as water boiled, power climbed further. A steam explosion blew the 1,000-ton lid off the reactor, and the graphite burned for days, carrying radioactive material across Europe. There was no Western-style containment building.

That last line is where the disputes live. The 4,000 figure is a projection using the linear no-threshold model for the most exposed groups. Some researchers and advocacy groups have published estimates of tens of thousands, or more, by applying the same model to the millions of people across Europe who received very small doses. The UN's scientific committee on radiation (UNSCEAR) has argued against multiplying tiny doses by huge populations to predict deaths, because the effect cannot be detected and the result depends entirely on the model. The WHO-led report also called the mental health impact, much of it from fear and relocation, "the largest public health problem unleashed by the accident."

Fukushima Daiichi, March 11, 2011

A magnitude 9.0 earthquake shut down the reactors as designed. The tsunami that followed, about 15 meters high at the site, flooded the backup diesel generators and electrical systems. Without power, cooling stopped. Decay heat melted the cores of Units 1, 2 and 3 within about three days, and hydrogen explosions wrecked reactor buildings. More than 100,000 people were evacuated; WNA cites figures up to about 160,000.

The World Nuclear Association, summarizing UNSCEAR's 2021 report, says no deaths or radiation sickness cases have been attributed to radiation from the accident, and future radiation-related health effects are "unlikely to be discernible." Japan's government did recognize, in 2018, a link between radiation and the lung cancer death of one plant worker. Separately, Japanese authorities counted 2,313 disaster-related deaths among Fukushima evacuees by September 2020, caused by the stress of evacuation, poor shelter conditions and interrupted medical care. About 90 percent were 66 or older.

That second number is the hard lesson of Fukushima. Some researchers have argued that the evacuation, especially of hospitals and nursing homes, killed more people than radiation would have if they had sheltered in place. Others answer that officials could not know the size of the release in the first days and had to act on the worst case. Both points belong in any fair account of what happened.

Deaths per terawatt-hour

To compare energy sources fairly, researchers divide deaths from accidents and air pollution by electricity produced. Our World in Data's widely cited version uses Markandya and Wilkinson (2007, The Lancet) for fossil fuels, Sovacool and colleagues (2016) for hydro, wind and solar, and its own calculation for nuclear that includes Chernobyl (433 deaths, a conservative count of direct and projected deaths) and Fukushima (2,314, using the Japanese government's figures).

Anti-nuclear groups argue these averages hide the real risk, which is a rare catastrophe that makes land unusable and costs hundreds of billions of dollars to clean up. That is a fair point about consequences and cost. It does not change the mortality record, and anyone weighing the issue needs both.

Radiation doses in everyday terms

U.S. agencies mostly use millirem. Most of the world uses millisieverts. Convert carefully: 1 millisievert (mSv) = 100 millirem (mrem). So 1 mrem = 0.01 mSv.

Half of the average American's yearly dose is natural, mostly radon. Nearly all of the other half is medical imaging. Normal operation of a nuclear plant is a rounding error in that total. (This is general education. Questions about your own medical imaging or radon at home are for your doctor or your state radon program.)

The linear no-threshold debate

Regulators set limits using the linear no-threshold model: risk is assumed to rise in a straight line with dose, all the way to zero. The NRC says it accepts LNT "as a conservative model," while noting that health data do not definitively show cancer from doses below about 10,000 mrem. Supporters, including most radiation protection bodies, say LNT is the prudent assumption when effects at low doses cannot be measured. Critics, including some radiation biologists and pro-nuclear groups, argue it overstates low-dose risk, drives costly over-regulation and fuels fear, which at Fukushima had its own death toll. Executive Order 14300 (May 2025) directed the NRC to reconsider its reliance on LNT and the related ALARA ("as low as reasonably achievable") standard. As of October 2026 the outcome of that review was still developing, and it is worth watching.