[en] Research on radiation carcinogenesis requires a twofold approach. Studies of primary molecular lesions and subsequent cytogenetic changes are essential, but they cannot at present provide numerical estimates of the risk of small doses of ionizing radiations. Such estimates require extrapolations from dose, time, and age dependences of tumor rates observed in animal studies and epidemiological investigations, and they necessitate the use of statistical methods that correct for competing risks. A brief survey is given of the historical roots of such methods, of the basic concepts and quantities which are required, and of the maximum likelihood estimates which can be derived for right censored and double censored data. Non-parametric and parametric models for the analysis of tumor rates and their time and dose dependences are explained. (author)