In a prospective, randomized clinical trial, we compared the accuracy of warfarin dosage-adjustments predictions using a computer program to the skill of an experienced anticoagulation nurse-specialist. The computer program predicts the steady state warfarin dose by applying Bayesian forecasting techniques to a mathematical model of the dynamic pharmacologic response to warfarin. Fifty patients who were receiving chronic warfarin therapy and who required a dosage adjustment because their prothrombin time was greater than or equal to 2 s away from their target prothrombin time were enrolled. The baseline characteristics of each group were similar, including the mean of the absolute value of the differences between initial prothrombin times and corresponding target prothrombin times. After a new a new warfarin dose was predicted, the prothrombin time was measured at least 7 days after dosage adjustment. Overall, the results in each group were comparable. There was no significant difference between groups and the mean of the absolute value of the differences between final prothrombin times and target prothrombin times, nor was there a difference in the proportion of patients who had a final prothrombin time within 2 s of the target prothrombin time. We conclude that the accuracy of warfarin dosage adjustments made using computer modeling is comparable to the skill of an anticoagulation nurse-specialist.