BACKGROUND AND PURPOSE The risk of stroke in patients with atrial fibrillation (AF) is not homogeneous, and various clinical risk factors have informed the development of stroke risk stratification schemes (RSS). Among anticoagulated cohorts, the emphasis should be on the identification of patients who remain at high risk for stroke despite anticoagulation. METHODS We investigated predictors of thromboembolism (TE) risk in an anticoagulated AF clinical trial cohort (n = 7329 subjects) and tested the predictive value of contemporary RSS in this cohort: CHADS₂, Framingham, NICE 2006, American College of Cardiology/American Heart Association/European Society of Cardiology 2006, the 8th American College of Chest Physicians guidelines and the CHA₂DS₂-VASc schemes. RESULTS On multivariate analysis, significant predictors of TE were stroke/TIA (hazard ratio [HR], 2.24; P <0.001), age 75 years or older (HR, 1.77; P = 0.0002), coronary artery disease (HR, 1.52; P = 0.0047), and smoking (HR, 2.10; P = 0.0005), whereas reported alcohol use (HR, 0.70; P = 0.02) was protective. Comparison of contemporary RSS demonstrated variable classification of AF patients into risk strata, although c-statistics for TE were broadly similar among the RSS tested and varied between 0.575 (NICE 2006) and 0.647 (CHA₂DS₂-VASc). CHA₂DS₂-VASc classified 94.2% as being at high risk, whereas most other RSS categorized two-thirds as being at high risk. Of the 184 TE events, 181 (98.4%) occurred in patients identified as being at high risk by the CHA₂DS₂-VASc schema. There was a stepwise increase in TE with increasing CHA₂DS₂-VASc score (P (trend) <0.0001), which had the highest HR (3.75) among the tested schemes. The negative predictive value (ie, the percent categorized as "not high risk" actually being free from TE) for CHA₂DS₂-VASc was 99.5%. CONCLUSIONS Coronary artery disease and smoking are additional risk factors for TE in anticoagulated AF patients, whereas alcohol use appears protective. Of the contemporary stroke RSS, the CHA₂DS₂-VASc scheme correctly identified the greatest proportion of AF patients at high risk, despite the similar predictive ability of most RSS evidenced by the c-statistic.