Is Nuclear Power Safe? What the Death-Rate Data Shows

Nuclear power is one of the safest ways to generate electricity, judged by the measure public health researchers use: deaths per unit of energy produced, counting both accidents and the ongoing toll of air pollution. That ranking survives Chernobyl and Fukushima being added to the total. The number is worth understanding, though, because of what it counts and what it leaves out.

The comparison researchers reach for is deaths per terawatt-hour, a unit large enough to make cross-fuel comparisons meaningful. One terawatt-hour is roughly what a mid-sized country’s grid delivers in a few hours, or what a single large power plant might produce in a year or two. The foundational version of this analysis came from Anil Markandya and Paul Wilkinson, who published a life-cycle comparison of electricity sources and their health effects in The Lancet in 2007. Benjamin Sovacool later built a separate historical database of energy accidents going back to 1907, and Our World in Data combines both lines of research into the numbers now cited everywhere from climate reports to op-eds.

What deaths per terawatt-hour counts

The figure isn’t just accident deaths. For fossil fuels, it’s mostly air pollution: the fine particulates and sulfur dioxide that coal and oil combustion push into the atmosphere, which raise rates of heart disease and stroke in the surrounding population year after year. Accidents, mine collapses and pipeline explosions among them, add a smaller layer on top.

By Our World in Data’s tally, coal kills about 24.6 people per terawatt-hour and oil about 18.4, almost entirely from that slow pollution burden rather than headline disasters. Gas comes in far lower, and nuclear lower still.

Source: Our World in Data, “What are the safest and cleanest sources of energy?”, synthesizing Markandya and Wilkinson (2007) with Sovacool’s accident database.

Hydropower’s low average hides a single huge outlier. Almost all of that death toll traces to one event: the Banqiao Dam failure in China in 1975, which killed an estimated 171,000 people. Take that one dam out and hydropower looks close to wind and solar. Leave it in, and it shows how a rare, severe event can dominate an average built from decades of data. Nuclear’s number carries the same kind of weight, just at a smaller scale.

Chernobyl and Fukushima, counted in

Nuclear’s 0.07 figure already includes both major reactor accidents. That’s the point of using a long historical average instead of pointing to how clean the fleet has looked since 2011.

Chernobyl, in 1986, killed 2 plant workers in the initial explosion and 28 more from acute radiation sickness in the following weeks, a toll confirmed within months. Separately, radioactive iodine released by the fire caused a spike in thyroid cancer among people who were children in the contaminated region at the time. By 2015, roughly 4,800 thyroid cancer cases had been linked to that exposure, and Our World in Data’s own accounting puts the most defensible estimate of the total death toll from Chernobyl at somewhere between 300 and 500 people, well below the World Health Organization’s 2005 upper-bound projection of 4,000 eventual cancer deaths and well above the 30 confirmed within weeks of the accident. The reactor involved, an RBMK design, lacked the containment structure standard on Western reactors and had a control-rod flaw that made an emergency shutdown briefly add reactivity instead of cutting it, a design gap World Nuclear Association’s own technical writeup treats as central to why the accident escalated the way it did.

Fukushima, in 2011, is a different story. The meltdown itself didn’t kill anyone directly. One plant worker died in 2018 from lung cancer that Japanese authorities attributed to radiation exposure on the job. The death toll instead comes from the evacuation: more than 150,000 people were moved out on short notice, and Japan’s government has certified 2,313 deaths as disaster-related, as of a September 2020 count, stemming from disrupted hospital care and the physical strain of relocating elderly and hospitalized patients under emergency conditions. Total confirmed deaths tied to Fukushima run to 2,314, and almost none of it came from radiation itself.

Both events also cost far more in property damage than in lives. Sovacool’s database, covering 279 major energy accidents between 1907 and 2007, found nuclear responsible for roughly 41% of the $41 billion in total damage across every energy source in that period, despite causing a small share of the deaths. Nuclear accidents are rare and, when they happen, expensive and disruptive on a scale fossil fuel accidents rarely reach. They’re just not what kills the most people, because they happen so infrequently against the scale of pollution that fossil fuels emit continuously.

Waste and radiation risk, in plain terms

Waste is the part of the nuclear safety conversation that the deaths-per-terawatt-hour number doesn’t touch at all, because it isn’t primarily a mortality question yet.

The volumes are smaller than most people expect. A 1,000-megawatt reactor, according to the World Nuclear Association, produces about 3 cubic meters of high-level waste a year when its fuel is reprocessed, against roughly 300,000 tonnes of ash from a coal plant generating a similar amount of power. Across all types, high and intermediate-level waste and used fuel make up less than 10% of total nuclear waste by volume, though it holds the large majority of the radioactivity. About 97% of used fuel can be recycled into new fuel. What can’t be recycled is vitrified, meaning it’s mixed into glass and sealed inside metal canisters, then buried in deep geological repositories, an approach several countries are now building out for permanent disposal rather than the interim storage most spent fuel sits in today.

Radioactivity also isn’t a fixed hazard. It decays. Used fuel’s radioactivity falls to about a thousandth of its initial level after 40 years. The World Nuclear Association states plainly that civilian nuclear waste has never caused a documented public death, a claim that holds up against records from decades of operation, though it says nothing about whether current storage arrangements will hold for the thousands of years some of this material stays hazardous. That’s a real open question, and it’s a different one from whether nuclear power kills people today.

From the operations side, this is where safety culture spends its time. Reactor incidents make headlines. Fuel handling and containment integrity are the daily, unglamorous work that keeps the incident rate low, and it’s the part of the job the public hears about least.

What this comparison does and doesn’t tell you

Deaths per terawatt-hour is a real, data-backed answer to a specific question: historically, which sources of electricity have killed the fewest people relative to how much power they’ve produced. On that question, nuclear does about as well as wind and solar, and vastly better than any fossil fuel.

It answers a narrower question than people often assume, though. It’s a global historical average, built mostly from 20th-century reactors and two specific accidents, not a forecast for a new plant built under current regulation. It treats a death from stress during a mass evacuation the same as a death from acute radiation exposure, which is defensible for counting purposes but erases a real difference in how those two things happen. It says nothing about the size of the area a bad accident makes uninhabitable, the decades a cleanup takes, or the cost of an accident when one does occur, all of which are much higher for nuclear than for the alternatives that also score well on this metric. And it doesn’t weigh in on cost overruns, construction timelines, or what to do with waste that outlasts the institutions built to guard it, which are the questions that decide whether new reactors get built.

None of that changes the mortality data. Being statistically very safe and having no downsides worth discussing are two different claims, and only the first one holds up under this particular number. As of 2023, nuclear supplied about 9% of the world’s electricity, a share the International Energy Agency expects to grow as countries weigh low-carbon options against the safety record the data shows.