Background: Since warfarin is primarily bound to serum albumin, hypoalbuminemia is likely to increase the free fraction of warfarin and to increase the risk of bleeding. We prospectively evaluated the impact of serum albumin levels (ALB) on international normalized ratio of prothrombin time (PT-INR) control and hemorrhagic events in atrial fibrillation (AF) patients treated with warfarin. Methods: Seven hundred fifty-five non-valvular AF patients on warfarin were enrolled. PT-INR control and major bleeding events (MB, International Society on Thrombosis and Haemostasis) were prospectively followed and were related to ALB at enrollment. Results: Twenty-seven patients developed MB during 1-year follow-up. In univariate/multivariate analyses, ALB (OR = 0.49, 95% CI 0.26-0.99, p = 0.04) and hemoglobin levels (OR = 0.78, 95% CI 0.65-0.92, p = < 0.01) were predictive for the annual risk of MB. In Spearman's rank correlation analysis, the baseline ALB was inversely correlated with the percentage of the time in PT-INR > 3.0 (ρ = -0.15, p < 0.0001), but neither 2.0 ≤ PT-INR ≤ 3.0 (ρ = 0.056, p = 0.13) nor PT-INR < 2.0 (ρ = -0.008, p = 0.82) during 1-year follow-up, suggesting that patients with low ALB had a directional tendency to be supratherapeutic control of PT-INR. The ROC curve showed that a cutoff of ALB was 3.6 g/dl to identify MB (AUC = 0.65). In Kaplan-Meier analysis, patients with ALB <3.6 g/dl (23/80, 29%) had more MB than those with ALB ≥3.6 g/dl (87/675, 13%, log-rank = 16.80, p < 0.0001) during long-term follow-up (3.8 ± 2.0 years). Conclusions: Hypoalbuminemia increases the likelihood of supratherapeutic PT-INR control and the risk of MB. ALB can be a practical surrogate marker to prevent excessive warfarin control and warfarin-related MB.