Cancers arise as the result of an accumulation of inherited and somatic mutations in proto-oncogenes and tumor-suppressor genes. In contrast to the activating mutations that generate oncogenic alleles from proto-oncogenes, tumor-suppressor genes are targeted by loss-of-function mutations in cancer cells. A third class of mutated genes, with a rather more indirect role in cancer initiation and progression, has been identified; namely, the DNA-repair-pathway genes. Their inactivation in cancer is presumed to contribute to the development of mutations in other genes that directly affect cell proliferation and survival, such as the oncogenes and tumor-suppressor genes. Because DNA-repair genes are affected by loss-of-function mutations in cancer, they are often considered to represent a subset of the tumor-suppressor genes.While a number of relatively straightforward approaches have enabled the identification of oncogenic alleles in cancer, identification of tumor-suppressor genes has proven more difficult. Somatic cell genetic studies provided early evidence that tumorigenicity was a recessive trait in many cancers. Based on such findings, the existence of tumor-suppressor genes was inferred. The somatic cell genetic approaches have provided a means to define specific chromosomal regions containing tumor-suppressor genes, even though few tumor-suppressor genes have been identified using the approaches.Knudson's epidemiologic studies of retinoblastoma led to a proposal that has subsequently been termed the "two-hit hypothesis." In brief, Knudson proposed that two inactivating mutations were necessary for retinoblastoma development. The first mutation at the retinoblastoma susceptibility locus could be either a germ line or somatic mutation, while the second mutation was always somatic. Knudson's hypothesis not only illustrated the mechanisms through which inherited and somatic mutations might collaborate in tumorigenesis, but it linked the notion of recessive genetic determinants for cancer susceptibility to the findings from the somatic cell genetic studies.To date, more than 20 tumor-suppressor genes have been localized and identified through various experimental approaches, often employed in concert. The approaches include cytogenetic studies of constitutional chromosomal alterations in cancer patients, linkage analyses to localize genes that predispose to cancer, and loss of heterozygosity (LOH) or allelic loss studies undertaken on matched pairs of normal and cancer tissue.The authenticity of a tumor-suppressor gene is most clearly established by the identification of inactivating germ line mutations that segregate with cancer predisposition, coupled with the identification of somatic mutations inactivating the wild-type allele in the cancers arising in those with a germ line mutation. Supportive, but less convincing, evidence of a tumor-suppressor role for other genes may be presented, such as the identification of somatic, inactivating mutations in a gene in one or more types of cancer, or its decreased or absent expression in cancers. Largely because of the difficulties in assigning causal significance to any gene solely based on somatic alterations in its sequence and/or expression in cancers, genes not affected by inactivating, germ line mutations are most appropriately considered as candidate tumor-suppressor genes until additional data are available.Powerful insights into the cellular functions of many tumor-suppressor proteins, such as pRb, p16, p53, and APC, have been obtained. It has become apparent that tumor-suppressor proteins function in a diverse array of signaling pathways and growth regulatory networks. In some cases, the protein products of tumor-suppressor genes and proto-oncogenes function in overlapping regulatory networks. Further studies of tumor-suppressor gene function will advance our understanding of cancer pathogenesis and the basis for the site-specific pattern of cancer often seen in individuals with germ line mutations in tumor-suppressor genes.