The use of computer-aided methods to predict the toxicity of drugs is described. These methods can assist in the identification of toxic compounds early in the drug development process. Thus, there is potential for these methods to be combined with combinatorial synthesis and library design. Quantitative structure-activity relationships allow for the prediction of individual endpoints, usually for restricted groups of compounds. Expert systems for toxicity prediction are based on a number of methodologies, each with its own strengths and weaknesses. The relative merit of each individual technique and methodology is described. However, more toxicity data are required, both to produce and to validate expert systems. Potential sources of new data include the use of high-throughput screening and microarrays for toxicology.