The failures at Japan's Fukushima Daiichi nuclear power plant after the tsunami on 11 March have led people to question the risks associated with nuclear power stations. But disasters on this scale can be considered to be very unlikely if it is assumed, like natural catastrophes, that the more energy that is required to cause an event, the less likely it is to happen.

In the biosphere, for example, extinctions of single species or communities of organisms occur much more frequently than mass extinctions. Such natural events show a logarithmic scale of frequency distribution when plotted against energy input.

Soon after the Chernobyl disaster, Kenneth Hsü suggested that man-made accidents might show a similar frequency–magnitude relationship (Nature 328, 22; 1987). He proposed that financial costs could be used as a measure of magnitude, because every small accident has a monetary equivalent — for example, in lost electricity-supply payments and insurance-claim payouts. Everything from the smallest mishaps (such as temporary shutdowns) to the largest catastrophes could thus provide a wide range of data points, putting statistical assessment of nuclear-accident probabilities on a firmer footing than, say, radiation-leak measurements (Nature 335, 391; 1988). It may be time to take a fresh look at Hsü's proposals for gauging the safety of nuclear power plants.