A series of multivariate statistical methods have been used to explore the results of a set of four in vitro short-term tests (STT) on 73 chemicals reported by the US National Toxicology Program (NTP). Cluster analysis showed that the mouse lymphoma mutation (MLY) and sister-chromatid exchange (SCE) were similar in performance, as were the Salmonella (STY) and chromosomal aberration test (CHA). The lack of association between tests using the same genetic end-point or at the same phylogenetic level found in previous analyses was confirmed in this study. Factor analysis was used to derive a scale of genetic damage. This measure was contrasted with rodent carcinogenicity; only a limited association was found (rank correlation coefficient, rs = 0.32). Linear discriminant analysis was used to study whether the STTs could be used to complement one another. The combination of STY with the other STTs did not improve significantly the prediction of rodent carcinogenicity of STY alone. In the entire set of chemicals, 33% were negative in STY and positive in at least two other STT, and 11% was negative in STY and positive in the three other tests. SCE and MLY were complementary to STY for identifying the most genotoxic chemicals, but CHA was not a useful complement. The presence of potential electrophilic sites in the chemicals was highly correlated with the STY results, but did not improve the ability of STY to identify genotoxic chemicals or predict rodent carcinogens. In conclusion, the other in vitro STTs did not complement STY for predicting carcinogenicity, but were an important complement for describing the potential genotoxicity of chemicals.