Abstract A conceptual scheme and associated statistical methodology is presented which is designed to provide the basis for a clinically useful computer program to suggest optimal dosage regimens for a number of drugs for individual patients. Routinely available clinical observations will be used to predict the parameters of a pharmacokinetic model, and subsequent blood level determinations used to refine these predictions, so that dosages can be designed to produce therapeutically desirable blood levels of agents. The system is intended to deal with most, if not all, of the influences on drug pharmacokinetics, to improve its performance by learning about the individual patient as well as the underlying population, and to modify its suggestions for an individual, should his clinical characteristics change. It may be used to discover new relationships between physiological states and drug pharmacokinetics. The system can exploit prior information in the form of theoretical and published data when its data base is small. An encouraging preliminary test of the system is reported.