Importance: Bleeding and thrombotic events (eg, stroke and systemic embolism) are common in patients with atrial fibrillation (AF) taking warfarin sodium despite a well-established therapeutic range. Objective: To evaluate whether history of therapeutic warfarin control in patients with AF is independently associated with subsequent bleeding or thrombotic events. Design, Setting, and Participants: In this multicenter cohort study of 176 primary care, cardiology, and electrophysiology clinics in the United States, data were obtained during 51 830 visits among 10 137 patients with AF in the Outcomes Registry for Better Informed Treatment of Atrial Fibrillation (ORBIT-AF) Registry; 5545 patients treated with warfarin were included in the bleeding analysis, and 5635 patients were included in the thrombotic event analysis. Patient follow-up was performed from June 29, 2010, to November 30, 2014. Data analysis was performed from August 4, 2016, to February 15, 2019. Exposures: Multiple measures of warfarin control within the preceding 6 months were analyzed: time in therapeutic range of 2.0 to 3.0, most recent international normalized ratio (INR), percentage of time that a patient had interpolated INR values less than 2.0 or greater than 3.0, INR variance, INR range, and percentage of INR values in therapeutic range. Main Outcomes and Measures: Association of INR measures, alone or in combination, with clinical factors and risk for thrombotic events and bleeding during the subsequent 6 months was assessed post hoc using logistic regression models. Results: A total of 5545 patients (mean [SD] age, 74.5 [9.8] years; 3184 [57.4%] male) with AF were included in the major bleeding analysis and 5635 patients (mean [SD] age, 74.5 [9.8] years; 3236 [57.4%] male) in the thrombotic event analysis. During a median follow-up of 1.5 years (interquartile range, 1.0-2.5 years), there were 339 major bleeds (6.1%) and 51 strokes (0.9%). Multiple metrics of warfarin control were individually associated with subsequent bleeding. After adjustment for clinical bleeding risk, 3 measures-time in therapeutic range (per 1-SD increase ≤55: adjusted odds ratio [aOR], 1.16; 95% CI, 1.02-1.32), variation in INR values (aOR, 1.32; 95% CI, 1.19-1.47), and maximum INR (aOR, 1.20; 95% CI, 1.10-1.31)-remained associated with bleeding risk. Adding INR variance to a clinical risk model slightly increased the C statistic from 0.68 to 0.69 and had a net reclassification improvement index of 0.028 (95% CI, -0.029 to 0.067). No INR measures were associated with subsequent stroke risk. Conclusions and Relevance: Three metrics of prior warfarin control were associated with bleeding risk but only marginally more so than traditional clinical factors. This study did not identify any measures of INR control that were significantly associated with stroke risk.