Knowledge of pharmacogenetics may help clinicians predict their patients' therapeutic dose of warfarin, thereby decreasing the risk of bleeding during warfarin initiation. Our goal was to use pharmacogenetics to develop an algorithm that uses genetic, clinical, and demographic factors to estimate the warfarin dose a priori. We collected a blood sample, demographic variables, laboratory values, smoking status, names of medications, and dietary history from 369 patients who were taking a maintenance dose of warfarin. Using polymerase chain reaction, we genotyped each participant for the presence of 8 polymorphisms in the cytochrome P450 2C9 system. Using multiple regression, we quantified the association between warfarin dose and all factors. Advanced age, lower body surface area (BSA), and the presence of cytochrome P450 2C9 *2 or *3 single nucleotide polymorphisms were strongly associated (P < 0.001) with lower warfarin dose: the maintenance dose decreased by 8% per decade of age, by 13% per standard deviation decrease in BSA, by 19% per 2C9*2 allele, and by 30% per 2C9*3 allele. Warfarin doses were 29% lower in patients who took amiodarone, 12% lower in patients who took simvastatin, 21% lower in patients whose target INR was 2.5 rather than 3.0, and 11% lower in white rather than African-American participants (P < 0.05 for these comparisons). An algorithm that included these factors and one of borderline significance (sex), explained 39% of the variance in the maintenance warfarin dose. Use of this pharmacogenetic model had potential to prevent patients from being overdosed when initiating warfarin: we estimate that only 24 (6.5%) patients would have been over- dosed by >2 mg/day with pharmacogenetic dosing compared to 59 (16%) patients who would have been overdosed if they had been prescribed the empirical dose of 5 mg/day (P < 0.001). In conclusion, the maintenance warfarin dose can be estimated from demographic, clinical, and pharmacogenetic factors that can be obtained at the time of warfarin initiation.