Toxicity testing has been ineffective in the prediction of drug candidates that will be associated with a relatively high incidence of idiosyncratic drug reactions (IDRs). Circumstantial evidence suggests the involvement of reactive metabolites in the aetiology of these reactions and this has prompted several companies to screen drug candidates for the formation of such compounds. Most drugs form at least one reactive metabolite. To develop efficient prediction methods, a better understanding of the basic mechanisms involved is essential. This review highlights the current mechanistic hypotheses of IDRs and discusses future directions in the development of better predictive tests.