Data from 511 cases of Wilms' tumor in France (including 12 familial cases) and 8 pedigrees from the literature were analyzed to test three modifications of Knudson's classical bimutational theory, based on genomic imprinting in Wilms' tumor carcinogenesis. Analysis of data of age at diagnosis and segregation analysis were performed to determine the number of independent events for Wilms' tumor development and to search for a differential role of paternal and maternal alleles. Unexpectedly, we show that only one rare event is required for tumor development in isolated unilateral cases which are considered to be mainly nonhereditary. In familial cases, we observe no effect of the sex of the transmitting parent on either hge at diagnosis or segregation ratio. We show that this could be explained by models of genomic imprinting which assume two nonindependent events, or only one rare genetic event. In bilateral cases we show a bimodality for age at diagnosis which could be due to a mixture of hereditary and nonhereditary cases. This result completely questions the classical assumption according to which all bilateral cases would be hereditary. These findings support the hypothesis that this childhood cancer arises from a variety of etiological pathways and might be useful to find strategies for further molecular investigations.