The analysis of the molecular mechanisms governing multistep carcinogenesis became experimentally approachable since the identification and characterization in tumor cells of altered or activated versions of cellular genes (oncogenes) that normally control cell growth and differentiation. The activating mutations confer new properties to the oncogene products and should therefore be considered as gain of function mutations. In addition, the oncogenes appear to act as dominant genetic traits since they act also in the presence of the homologous wild-type allele. However, the concept of a dominance of the transformed phenotype has been challenged by early experiments with somatic cell hybrids which showed that the fusion of normal and malignant cells may suppress the tumorigenic phenotype. The suppression or reversion of the malignant phenotype by the introduction of a normal chromosome into a tumor cell line has lent support to the idea that a family of cellular genes are coding for factors capable to interact with the cell-growth control machinery. These genes seem to reconstitute the normal control of cell growth even in the presence of an activated oncogene. In addition, a two-mutation model has been proposed to explain the epidemiological and clinical features of childhood cancers. According to the model, the development of these malignancies can be caused by the loss or inactivation of both alleles of cellular genes, as suggested by the somatic cell hybrid experiments where the function of the inactivated genes is restored by the contribution of those derived from the normal parental cells. This family of genes is designated as onco-suppressor genes since their product is necessary for the normal regulated cell growth and is lacking or inactivated in malignant cells. At gene level they should be considered as recessive genetic traits, since the tumor phenotype appears when both alleles of an onco-suppressor gene are inactivated. The mutations affecting their normal functions belong to the type "loss of function". The molecular analysis of retinoblastoma has led to the cloning and sequencing of the related onco-suppressor gene (RB gene) whose product displays the features of a gene-regulatory protein. In addition, a binding between the RB product and various viral onco-proteins (E1A, large T, E7) has been demonstrated, thus suggesting a mechanism of RB inactivation by which some DNA viruses can transform the host cell.(ABSTRACT TRUNCATED AT 400 WORDS)