Krypton-85 85Kr is a beta emitter with 687keV maximum energy and 10.7 year half-life. It's produced by uranium fission with 0.27% yield (137Cs and 90Sr: 6%). Here is a report, a bit old and in French, but figures read the same in English anyway: http://www.irsn.fr/EN/Research/publications-documentation/radionuclides-sheets/Documents/Kr85SAN.pdf When reprocessing used nuclear fuel rods at La Hague, solid fission products like 137Cs and 90Sr are separated for storage, but 85Kr is just diluted and emitted in the atmosphere. For instance in 1999, La Hague emitted 2.9*1017 Bq of 85Kr, more than the radioactivity of 137Cs and 90Sr released by the Chernobyl disaster. Iodine, caesium, strontium radioisotopes fall on the soil more quickly and locally if emitted, and accumulate in the food chain. Krypton, as a noble gas, dilutes in the whole northern hemisphere's atmosphere, where in 2001 it added 1.2 Bq/m3, with 3/4 of it coming from La Hague - it must be worse by now. How bad is such an added radioactivity, much smaller than natural sources anyway? It irradiates a human body by 4nSv/yr, or 500nSv/yr at the skin as beta rays don't go deep. No experimental data tells the effect of such a low dose; scientific near-consensus is to extrapolate proportionally from higher doses, taking 1% more risk of a fatal cancer per 0.2Sv exposure. Over 30 years, this adds only 6*10-9 risks of casualty - but for each of the 6 billion people in the hemisphere, meaning an estimated 36 casualties. Note the 1% per 0.2 Sv is a whole-body average, but external beta rays damage essentially the skin. It seems that 500 nSv/yr at the skin is a bit worse than the deep 4 nSv/yr taken here. Some people challenge that the risk is proportional to the dose; they want to see some dose threshold below which the risk vanishes. Among them is Areva, the operator of La Hague. I consider that since 85Kr arrives from the power plants confined in fuel rods, keeping all 85Kr confined in some storage can't be that difficult. And to avoid three dozens possible deaths, it should be done, without arguing about risk models.