RÉSUMÉLes études épidémiologiques concernant des personnes exposées aux rayonnements ionisants restent à la base des estimations de risque de cancer radioinduit chez l'homme. La principale étude ayant servi à l'élaboration des normes de radioprotection est celle des survivants d'Hiroshima et Nagasaki. These epidemiological studies allow to confirm that a risk does exist for some types of cancer following high-dose exposures often at high dose-rates. However, no conclusion can be drawn for low doses and low dose-rates. Therefore we have to extrapolate from known high-dose risks to low doses and low dose-rates by various dose-response patterns. Another difficulty in assessing radiation cancer risks comes from the long latency time, which explains that all excess cancers have not yet been observed in the irradiated population studied.Once more, mathematical models are used to project excess lifetime cancer mortality. The estimations of radiation cancer risks are therefore marked by a great number of uncertainties, since they change accordingly to the model used. Other uncertainties come from the data, especially the dose estimates and are heightened when extrapolating to other populations.In 1988, UNSCEAR assessed new estimates for excess lifetime cancer mortality in the range of 4 to 11 % per gray. These values mean a revaluation of the previous estimates by a 1.6 to 4.4 factor, which is mainly consecutive to the use of different projection models. Besides, they are solely based on the Hiroshima and Nagasaki survivors.whereas patient studies assess a lower risk. Finally UNSCEAR does not precisely state what is the available reduction factor to modify risks for low doses and low dose rates which should lie between 2 and 10. Due to a number of persistant uncertainties, we should not consider it justified to revise protection standards presently.