A well characterized model of multistage carcinogenesis is that of hepatocarcinogenesis in the rat. The histopathology as well as the cell and molecular biology of the stages of initiation, promotion, and progression have been elucidated to varying degrees in this system. Putatively single initiated hepatocytes are identified by their expression of the ubiquitous marker of hepatocarcinogenesis, glutathione-S-transferase pi (GSTP). 0.5-1.0 x 10(6) GSTP-positive "initiated" hepatocytes developed within 14 days after initiation with a subcarcinogenic dose of diethylnitrosamine (DEN). Approximately 1% of these cells develop clonally into altered hepatic foci (AHF) in animals administered promoting agents, such as phenobarbital, chronically for 4-8 mo. Hepatocytes within AHF during the stage of promotion exhibit normal diploid karyotypes but various phenotypes depending on the chemical nature of the promoting agent. Continued administration of the promoting agent results in the infrequent development of hepatocellular carcinomas; however, administration of a complete carcinogen or a progressor agent during the stage of promotion results in substantial numbers of hepatic neoplasms. In order to quantitate the development of the stage of progression more accurately, markers selective for this stage have been sought. Transforming growth factor-alpha (TGF-alpha) appears to be such a marker of progression. About 500 TGF-alpha-positive lesions develop spontaneously following initiation and continued promotion, usually within GSTP-positive AHF, but administration of a single dose of a progressor agent such as ethylnitrosourea may increase this number 3-fold or more. Some agents such as gamma radiation and hydroxyurea, when administered as single or a few closely spaced multiple doses, result in no increased number in TGF-alpha-positive lesions but a markedly enhanced increase in their growth rate. By monitoring gene expression using quantitative stereology, the stages of hepatocarcinogenesis can be analyzed and quantified in sufficient detail so that the animal data can be utilized in biomathematical modeling to develop more accurate models for estimation of human cancer risks.