Introduction: In randomized trials of patients with non-valvular atrial fibrillation (AF), new oral anticoagulants (NOACs) have been shown to confer a lower risk of major bleeding than warfarin. Outside of these select study settings, bleeding outcomes for these medications have not been examined sufficiently. We studied a national sample of patients with AF to explore bleeding risks associated with NOACs in routine clinical settings. Methods: Using a 40% Medicare random sample denominator file and associated inpatient, outpatient and prescription administrative data from 2009-2012, we identified patients age 65 and older with new onset, non-valvular AF. Cox regression models estimated risk of upper gastrointestinal (UGI) and nontraumatic intracranial (IC) bleeding associated with NOAC (dabigatran and rivaroxaban) compared to warfarin use, adjusting for individual characteristics, morbidities, and time-varying medication use. The analysis excluded person-months with concurrent prescriptions filled for an antiplatelet medication (e.g. clopidogrel). Results: We identified 157,390 incident AF patients, mean age 77.8 (standard deviation (SD) 7.8); mean follow up was 10.8 months (SD 6.9); In the first year post-index diagnosis, 41.5% of patients received some oral anticoagulant; of these, 28.2% used NOACs (75% dabigatran, 25% rivaroxaban). Overall 24,661 bleeding events were observed (11,015 UGI and 1,868 IC). Out of 1,487,675 person-months, 212,670 (14.3%) were excluded due to concurrent prescription antiplatelet exposure. Compared to warfarin, the hazard ratio for bleeding associated with NOAC was 1.33 (95% CI 1.21 to 1.47) for UGI, 0.59 (95% CI 0.42 to 0.82) for IC, and 1.09 (95% CI 1.02, 1.17) for all significant bleeding combined. Conclusions: Compared with warfarin, NOACs were associated with a lower risk of IC bleeding, a higher risk of UGI bleeding and a modestly higher risk of significant bleeding overall. These results should inform clinical decision-making and selection of individualized thromboembolism prophylaxis strategies among AF patients. Further studies are needed to illuminate the divergent UGI and IC bleeding risks observed and to identify patient-level factors predictive of these outcomes.