OBJECTIVES: To examine the relationship between International Normalized Ratio (INR) and outcomes (major bleeding events and strokes) in atrial fibrillation (AF) patients on anticoagulation with warfarin. METHODS: A systematic review and meta-analysis of studies published in English between January 1, 1985 and October 30, 2002 was performed. MEDLINE (PubMed), Current Contents, and relevant reference lists were searched. Studies enrolling patients with nonvalvular AF on warfarin anticoagulation were eligible for inclusion if they reported stroke and/or major bleeding events in relation to INR, or time spent in therapeutic range. The risk of bleeds in overanticoagulated patients (INR>3) and the risk of strokes in underanticoagulated patients (INR<2) was assessed. RESULTS: Twenty-one studies (6,248 patients) met all inclusion criteria. Of the 21 studies, target conventional INR of 2 to 3 was used in 9. An INR < 2, compared with an INR > 2, was associated with an odds ratio (OR) for ischemic events of 5.07 (95% confidence interval (CI) = 2.92, 8.80). An INR > 3, compared with an INR < 3, was associated with an OR for bleeding events of 3.21 (95% CI = 1.24, 8.28). On average, in the four studies with a target INR range of 2 to 3, AF patients on warfarin spent 61% of time within, 13% of time above and 26% below the therapeutic range. CONCLUSION: Available evidence indicates that in patients with non-valvular AF, the risk of ischemic stroke with insufficient warfarin anticoagulation (INR < 2), and the risk of bleeding events with overanticoagulation (INR > 3) is significantly higher relative to AF patients maintained within the recommended INR of 2 to 3. However, the data are sparse, heterogeneous, and mostly based on clinical trials. More studies evaluating clinical outcomes in relation to INR are needed, especially in a real-world setting.