Global gene expression profiles in cancer have impacted both classification of tumors and definition of molecular pathways in neoplasia. To explore the possibility of employing human tumor cell lines to obtain information on the functional genomics of the early stages of tumorigenesis, we have characterized variation in gene-expression patterns in a cytogenetically well-defined series of cell lines derived from human hepatocellular carcinoma (HCC). Microarrays containing 6,720 sequence-verified human cDNAs were used in this study. Nineteen well-characterized HCC cell lines were analyzed, and a nontumorigenic liver-derived epithelial cell line (Chang) was used as a reference. Each sample was examined at least twice by switching fluorescent dyes, Cy-5 and Cy-3, and average values of 2 experiments on each sample were used for further analysis. Analysis of the clustered data revealed 2 distinctive subtypes of gene-expression patterns among the 19 cell lines, suggesting a degree of heterogeneity among the gene-expression profiles of cell lines. Remarkably, expression of alpha-fetoprotein (AFP) was highly correlated with the molecular subtypes of HCC. Although the 3 most distinctive gene-expression modules represented the signatures of 2 different subgroups of HCC, most of the cell lines shared many coexpressed genes. However, sets of coexpressed genes that are specific for the subtypes of HCC were identified. Furthermore, our results indicate that the comparison between gene-expression patterns and structural alterations in chromosomes is potentially useful in identifying genes critical in early stages of tumorigenesis. In conclusion, these results not only identified unrecognized subtypes of HCC, but also provided potential molecular markers for each subtype that can be useful for diagnostic and/or therapeutic purposes.