Non-animal procedures, including in vitro test systems and test strategies, can already make a significant contribution to the background to risk assessment — in predicting both the toxic potential and toxic potency of chemicals, as well as, in some circumstances, the toxic hazard they may represent under specified conditions of exposure. They can be particularly useful for investigating molecular and cellular mechanisms of chemical-induced toxicity, and for identifying species-specific effects, which greatly limit the value of data from laboratory animal studies in the human risk assessment process. Attention is focused on the need for greater effort to be invested in the development of non-animal procedures for evaluating the biokinetic factors which will determine the ultimate form and concentration of a particular chemical at possible sites of toxic action. The relative merits of correlative and mechanistic approaches to test development and test validation are discussed. The need for realism is emphasised, not only in relation to our expectations of the validation process, but also in terms of the current and future status of regulatory toxicology, in vitro or in vivo, as a scientific discipline. Finally, it is concluded that the intelligent and strategic use of in vitro test systems, in conjunction with predictive computer modelling, could markedly improve the scientific basis of human risk assessment.