BACKGROUND: The predictive value of bleeding risk scores for atrial fibrillation in older patients is not as well known. The goal of this study was to evaluate the predictive value of HASBLED, ORBIT and ATRIA for major bleeding (MB) and intracranial hemorrhage (ICH) in patients ≥ 75 years with atrial fibrillation and oral anticoagulation (OAC). METHODS: A retrospective unicenter study including patients ≥ 75 years with atrial fibrillation (AF) and OAC. A total of 7613 patients ≥ 75 years with AF and OAC included between 2014 and 2018 (registry: NCT04364516). We analyzed the discriminative value of HASBLED, ATRIA and ORBIT scores for bleeding endpoints (major bleeding as primary endpoint and intracerebral hemorrhage as secondary). Cox regression was used to predict major bleeding with each scale and also for searching other variables potentially predictor of major bleeding. Model discrimination was assessed using Harrell's C-statistic. Calibration was assessed with goodness-of-fit test proposed by Gronnesby and Borgan. RESULTS: During a mean follow up of 4.0 years (IQR: 2.4-5.7 years), 729 patients developed MB (2.61 per 100 patients/year) and 243 patients developed ICH (0.85 per 100 patients/year). Three scores showed a low discrimination for major bleeding, being ORBIT the best (HASBLED C statistic = 0.557; ATRIA C statistic = 0.568; ORBIT C statistic = 0.595) and also a low discrimination for ICH (HASBLED C statistic = 0.509; ATRIA C statistic = 0.522; ORBIT C statistic = 0.526). Among the variables that are part of the scores and other baseline characteristics, after multivariable adjustment only sex (male), dementia, prior admission for bleeding, anemia and liver disease were found as a predictors of MB. CONCLUSIONS In older patients under oral anticoagulation with atrial fibrillation, the risk scores HASBLED, ATRIA and ORBIT showed a weak discrimination for major bleeding and intracranial hemorrhage. Therefore, other better alternatives should be evaluated for this purpose.