[en] The methods employed in evaluating radiological risk are based, by and large, on the non-threshold linear relationships proposed by the ICRP and the Advisory Committee on the Biological Effects of Ionizing Radiations (BEIR) of the US Academy of Sciences in its 1972 report. Recently, the contentious discussions surrounding the latest BEIR report and the publication of new estimates of the doses received at Hiroshima and Nagasaki have reopened the controversy surrounding dose-effect relationships. The chief point at issue is the shape of this relationship - whether or not it is linear and whether it has a threshold - and the role of radiation in carcinogenesis, both in association with other risk factors and as an independent factor. The present paper leaves the theoretical questions aside and instead shows how, by marshalling information on irradiated human populations, a sequence of calculations can be set up to enable specialists to assign numerical values to the coefficients for these relationships. In an empirical approach such as this, theoretical hypotheses are no longer the sole factors at work. Comparatively pragmatic choices, such as the sample on the basis of which the curves are adjusted, the survival tables used, or even the simplifications made in taking account of sex and age, can also influence the result obtained. Sensitivity analyses are presented for the value of the coefficient of cancer induction per unit dose, showing that controversial theoretical options such as the transformation of a linear into a quadratic linear relationship have no greater effect on the result than other more contingent choices. Lastly, the implications of these sensitivity analyses for the practical application of dose-risk relationships are examined in relation to impact studies, optimization procedures, etc. (author)