Health records of the Japanese bomb survivor population [with the 1965 (T65D) and 1986 (DS86) dosimetry systems] have been analyzed and some evidence found for the selection effect hypothesized by Stewart and Kneale. This is found to be significant in only the first of the periods examined (1950-1958), and the effect diminishes in magnitude thereafter. There are indications that the effect might be an artifact of the T65D dosimetry, in which it is observed more strongly than in the DS86 data. There is no evidence to suggest that selection on this basis might confer correspondingly reduced susceptibility to radiation-induced cancer. If, however, one makes this assumption, as suggested by Stewart and Kneale, then current estimates of population cancer risks might need to be inflated by between 5% and 35% (for excess cancer deaths, Gy-1) or between 8% and 40% (for years of life lost, Gy-1) to account for this. It is likely that these figures, even assuming them not to be simply an artifact of the T65D dosimetry, overestimate the degree of adjustment required to the risk estimates.