Quantitative two-dimensional gel electrophoresis was used to compare the cellular protein patterns of a normal foreskin-derived human fibroblast cell line (LG1) and three immortal derivatives of LG1. One derivative, designated MSU-1.1 VO, was selected for its ability to grow in the absence of serum and is non-tumorigenic in athymic mice. The other two strains were selected for focus-formation following transfection with either Ha-ras or N-ras oncogenes and form high grade malignant tumors. Correspondence and cluster analysis provided a nonbiased estimate of the relative similarity of the different two-dimensional patterns. These techniques separated the gel patterns into three distinct classes: LG1, MSU-1.1 VO, and the ras-transformed cell strains. The MSU-1.1 VO cells were more closely related to the parental LG1 than to the ras-transformed cells. The differences between the three classes were primarily quantitative in nature: 16 % of the spots demonstrated statistically significant changes (P < 0.01, T test, mean ratio of intensity > 2) in the rate of incorporation of radioactive amino acids. The patterns from the two ras-transformed cell strains were similar, and variations in the expression of proteins that occurred between the separate experiments obscured consistent differences between the Ha-ras and N-ras transformed cells. However, while only 9 out of 758 spots were classified as different (1 %), correspondence analysis could consistently separate the two ras transformants. One of these spots was five times more intense in the Ha-ras transformed cells than the N-ras. These results demonstrate the potential of automated two-dimensional gel analysis for classifying different gel patterns, even patterns whose differences are obscured by the minor changes in spot intensity that arise between separate cell cultures.