Background Oral anticoagulation is the mainstay of stroke prevention in atrial fibrillation (AF), but must be balanced against the associated bleeding risk. Several risk scores have been proposed for prediction of bleeding events in patients with AF. Objectives To compare the performance of contemporary clinical bleeding risk scores in 18 113 patients with AF randomized to dabigatran 110 mg, 150 mg or warfarin in the RE-LY trial. Methods HAS-BLED, ORBIT, ATRIA and HEMORR2HAGES bleeding risk scores were calculated based on clinical information at baseline. All major bleeding events were centrally adjudicated. Results There were 1182 (6.5%) major bleeding events during a median follow-up of 2.0 years. For all the four schemes, high-risk subgroups had higher risk of major bleeding (all P < 0.001). The ORBIT score showed the best discrimination with c-indices of 0.66, 0.66 and 0.62, respectively, for major, life-threatening and intracranial bleeding, which were significantly better than for the HAS-BLED score (difference in c-indices: 0.050, 0.053 and 0.048, respectively, all P < 0.05). The ORBIT score also showed the best calibration compared with previous data. Significant treatment interactions between the bleeding scores and the risk of major bleeding with dabigatran 150 mg BD versus warfarin were found for the ORBIT (P = 0.0019), ATRIA (P < 0.001) and HEMORR2HAGES (P < 0.001) scores. HAS-BLED score showed a nonsignificant trend for interaction (P = 0.0607). Conclusions Amongst the current clinical bleeding risk scores, the ORBIT score demonstrated the best discrimination and calibration. All the scores demonstrated, to a variable extent, an interaction with bleeding risk associated with dabigatran or warfarin.