Environmental contaminants with common mode of toxic action (MOA) are generally expected to have similar structures andyor physico-chemical properties. Calculated descriptors of lipophilic, electronic and steric properties were used to cluster 115 test chemicals by MOA into nine different toxicant classes (non-polar non-speci®c toxicants, polar non-speci®c toxicants, uncouplers of oxidative phosphorylation, inhibitors of photosynthesis, inhibitors of acetylcholinesterase, inhibitors of respiration, thiol-alkylating agents, reactives (irritants), estrogenic compounds).
Stepwise discriminant analysis of the test chemicals resulted in 89.6% correct classi®cations into the MOA classes. The ®nal model uses 10 signi®cant variables (log K OW , e HOMO , V , Q AV , H MAX , MR, MW, D EFF , SASA, SAVOL). PLS discriminant analysis of the same data set resulted in a three-component model with r 0.89; the variables with the highest discriminatory power are log K OW , H MAX , D EFF and Q AV . Each MOA class reveals a characteristic pro®le in physico-chemical properties. Deviations relative to non-speci®c baseline toxicants are speci®c for each MOA class and re¯ect the structural dependences of the rate-limiting interactions that are causing the respective toxicities (functional similarity). By combining physiological and chemical knowledge about underlying processes, it is possible to indicate descriptor-based discrimination criteria by MOA as an essential prerequisite for rational selection and application of process-related QSARS for predictive purposes.