Germ-line mutations (present in all cells) in genes that are crucial for the cell cycle cause cancer only in specific cell lines (e.g. mismatch repair genes in the colon; BRCA1-2 in breast and ovary; other cancers in Bloom syndrome, neurofibromatosis and xeroderma pigmentosum). The mutation rate of genes other than mismatch repair or p53 is the same in colon cancer and in normal cells, indicating that a 'mutator phenotype', increasing the rate of mutations in many genes, is not an essential feature of sporadic cancers; conversely, fusion genes, TEL-AML1/AML1-ETO, typical of leukemia, are 100 times more frequent at birth than in overt leukemia in children, indicating that further selective events are needed to cause malignancy. The devastating impairment of immunity, as in AIDS patients, does not cause cancer other than Kaposi's sarcoma and non-Hodgkin's lymphoma, although immunological control is considered to be an essential mechanism in preventing the spread of cancer cells. These observations suggest that cell-specific additional events are needed to explain carcinogenesis. Carcinogenesis has been traditionally interpreted as the sequence of initiation (mutation) and promotion (clone expansion), with an interesting similarity with the neo-Darwinian theory of evolution, based on a first stage of genetic change (including recombination) and a second stage of selection. I propose that carcinogenesis consists in two general phases (not necessarily stages), i.e. genetic change followed by clone expansion (selective advantage). As in neo-Darwinian theory selection is chiefly represented by the elimination of the less fit, the selection of mutated cells would mainly consist in resistance to apoptosis or other types of 'bottlenecks' that hamper a cell's survival; an example of such a bottleneck is the autoimmunity that induces paroxysmal nocturnal hemoglobinuria in individuals with PIG-A mutations. Cancer rates show great variation in different countries around the world, a variation only marginally explained by genetic differences. More interestingly, migrants change their risk of cancer by adapting to that of the population into which they move: as these changes are not likely to be entirely due to mutagens in the environment, we have to invoke selective pressure over mutated cells to explain them. My theory is that mutated cells adapt to environmental 'niches' better than normal cells, in a 'gene-environment interaction' that involves the history of the genetic changes the cell has undergone and the kind of environment in which it happens to live.