Objective The importance of pharmacogenetic algorithms in warfarin dose prediction has drawn more attention. However, their performance has still been influenced by race and multiple clinical factors including warfarin dosage. The aim of this study was to evaluate the relationship of warfarin dosage with polymorphisms of CYP2C9*3, VKORC1-1639 A/G and CYP4F2*3, and the accuracy of pharmacogenetic algorithms. Methods We detected the polymorphisms of CYP2C9*3,VKORC1-1639 A/G and CYP4F2*3 in a cohort of Chinese patients (n＝282) under low intensity warfarin anticoagulation with target international normalized ratio (INR) ranging from 1.6 to 2.5, and evaluated the relationship between warfarin dosage and these polymorphisms. Then, the patients were stratified to 3 groups according to the dose range: higher dose group (≥4.5 mg/d), intermediate dose group (1.5～4.5 mg/d), and lower dose group (≤1.5 mg/d). We evaluated the performances of 8 eligible pharmacogenetic algorithms in different dose range. The performance of each algorithm was evaluated by calculating the percentage of patients whose predicted dose fell within 20% of their actual therapeutic dose (percentage within 20%), and the mean absolute error (MAE) between each predicted dose and actual stable dose. Results In the entire cohort, warfarin doses differed significantly across genotypes for CYP2C9*3 (P＜0.001), VKORC1-1639 A/G (P＜0.001), and CYP4F2*3 (P＝0.025). The algorithms from Caucasian and racially mixed populations tended to perform better in higher dose group, and algorithms from Asian populations performed better in intermediate dose group. None of the algorithms performed well in lower dose group. Conclusions Polymorphisms of CYP2C9*3, VKORC1-1639 A/G and CYP4F2*3 associate with stable warfarin dose in Chinese patients. No eligible pharmacogenetic algorithm could perform satisfactorily for all dosing range in the Chinese patients.