AbstractCancer is naturally very common, and practical questions about the possibility of radiation-induced harm are often questions about what in other contexts would be called background noise. Central to the question of whether small radiation exposures are carcinogenic is the effect of antenatal radiography. A comparison of singleton and twin births with radiography rates of 10 and 55 per cent respectively showed that radiography must be the main cause of the elevated frequency of malignant disease. In Japanese bomb survivors, most radiation-induced cancer has been found in those irradiated in adult life, less in those irradiated in childhood and adolescence, and least for exposurein utero. Specific biological differences between different kinds of malignant disease in their induction by ionizing radiation are becoming increasingly evident. When dose-response relationships for observed cancer frequencies are to be used as evidence about dose-response relationships for cancer induction, it will always be necessary to allow for the concomitant cell sterilization. When this is done, there is little support for linearity as the method of extrapolation when making predictions about possible effects of low doses but the absence of threshold seems scientifically inescapable. In cellular terms, radiation induction of cancer must be a very rare phenomenon, so rare compared with cell sterilization or mutation induction, that the general corpus of radiobiological understanding may be inapplicable.