BACKGROUND: Warfarin sodium is commonly prescribed for the prophylaxis and treatment of venous thromboembolism. Dosing algorithms have not been widely adopted because they require a fixed initial warfarin dose (eg, 5 mg) and are not tailored to other factors that may affect the international normalized ratio (INR). OBJECTIVE: TO develop an algorithm that could predict a therapeutic warfarin dose based on drug interactions, INR response after the initial warfarin doses, and other clinical factors. METHODS: We used stepwise regression to quantify the relationship between these factors in patients beginning prophylactic warfarin therapy immediately prior to joint replacement. In the derivation cohort (n = 271), we separately modeled the therapeutic dose after 2 and 3 initial doses. We prospectively validated these 2 models in an independent cohort (n = 105). RESULTS: About half of the therapeutic dose variability was predictable after 3 days of therapy: R 2 was 53% in the derivation cohort and 42% in the validation cohort. INR response after 3 warfarin doses (INR 3) inversely correlated with therapeutic dose (p < 0.001). Intraoperative blood loss transiently, but significantly, elevated the postoperative INR values. Other significant (p < 0.03) predictors were the first and second warfarin doses (+7% and +6%. respectively, per 1 mg), and statin use (-15.0%). The model derived after 2 warfarin doses explained 32% of the variability in therapeutic dose. CONCLUSIONS: We developed and validated algorithms that estimate therapeutic warfarin doses based on clinical factors and INR response available after 2-3 days of warfarin therapy. The algorithms are implemented online at www.WarfarinDosing.org.