The simplest complete system accounting for the time-course of changes in the prothrombin time induced by warfarin requires the combination of 4 independent models: A pharmacokinetic model for the absorption, distribution, and elimination of warfarin. Warfarin is essentially completely absorbed, reaching a maximum plasma concentration between 2 and 6 hours. It distributes into a small volume of distribution (10 L/70kg) and is eliminated by hepatic metabolism with a very small clearance (0.2 L/h/70kg). The elimination half-life is about 35 hours. A pharmacodynamic model for the effect of warfarin on the synthesis of clotting factors (prothrombin complex). Prothrombin complex synthesis is inhibited 50% at a warfarin concentration of about 1.5 mg/L. Warfarin concentrations associated with therapeutic anticoagulation are of similar magnitude. A physiological model for the synthesis and degradation of the prothrombin complex. The synthesis rate is about 5%/h/70kg and the elimination half-life estimated from changes in prothrombin time is approximately 17 hours. On average it will take 3 days for the anticoagulant effect of warfarin to reach a stable value when warfarin concentrations are constant. A model for the relationship between the activity of prothrombin complex and the prothrombin time. In general there is a hyperbolic relationship between these quantities. Its exact shape depends upon the method used for measuring the prothrombin time. Attempts to integrate these models into a single system have used essentially the same pharmacokinetic, physiological, and prothrombin activity models. Four distinct pharmacodynamic models have been proposed: linear, log-linear, power and Emax. One might be preferred on theoretical grounds (Emax) but its performance is not clearly different from the others. Empirical methods for warfarin dose prediction as well as those based on the combined pharmacokinetic-pharmacodynamic-physiological-prothrombin system have been proposed. Only one (which was also the first) [Sheiner 1969] has been adequately described and compared with the performance of an unaided physician. The programme compared favourably with decisions made by those physicians normally responsible for adjusting warfarin dose, but was not tested prospectively. A sizeable body of theoretical and experimental observations has contributed to our understanding of the warfarin dose-effect relationship. It remains to be demonstrated that any alternative method is superior to the traditional empirical approach to warfarin dose adjustment.