A major obstacle to the use of short‐term test results for predicting carcinogenicity is the nonavailability of methods that generate unequivocal estimations of potential carcinogenicity, even in the case of contrasting test results. We compared different mathematical classification methods such as linear discriminant analysis, theK‐nearest neighbor rule (KNN), and a new ad hoc classification method called dynamic carcinogenicity assessment (DYCA). The DYCA was developed by us to explicitly reflect the biological results. Through an analysis of a real data base we pointed out the advantages and limitations of these methods. KNN was clearly inferior to the other two algorithms. The novel DYCA approach was more sensitive to carcinogens, whereas linear discriminant analysis proved better for the identification of noncarcinogens.