Abstract Background Due to the fear of increased risk of bleeding, anticoagulation treatment is underutilized in the prevention of stroke in elderly patients with non-valvular atrial fibrillation (NVAF). Although direct oral anticoagulants (DOAC) are safer than VKA, still little is known about the risk factors associated with bleeding in elderly patients treated with DOAC. Furthermore, it is still uncertain whether the risk scores that are currently used can serve to effectively identify higher bleeding risk in elderly subjects. Purpose The aim of this study was to identify predictors of bleeding in a cohort of elderly people affected by NVAF treated with DOAC, and to evaluate the accuracy of risk scores for bleeding used at present. Methods Data on outpatients aged ≥75 years, naïve for DOAC therapy, who started therapy with Dabigatran, Rivaroxaban, Apixaban or Edoxaban for the prevention of thromboembolism during FANV were analyzed. HASBLED, ATRIA, OBRI and ORBIT scores were calculated for each patient. Patients had follow-up for 12 months during which deaths, therapy discontinuation and adverse events such as thromboembolism and bleeding were reported. Potential predictors of bleeding and the predictive value of each bleeding score were tested using univariate Cox regression; testing accuracy was evaluated using ROC curves. Results A total of 291 patients (52.9% female, mean age 82.85±5.18 years) had a median follow-up time of 11 (10–12) months. The incidence rate of major bleeding was 4.7 per 100 patient-years, the rate of intracranial bleeding was 0.4 per 100 patient-years. Patients who had major bleeding were more often affected by heart failure (63.6% vs 25%; p=0.009) and thrombocytopenia (36.4% vs 7.4%; p=0,009). However in the multivariate analysis only heart failure remained statistically associated with major bleeding (HR 3.83, 95% CI 1.06–13.85; p=0.041). None of tested bleeding risk scores was able to predict major bleeding in our cohort. HASBLED and ORBIT scores were able to predict major and non-major clinically relevant bleeding (HR 1.32; 95% CI 1.01–1.71; p=0.042 and HR 1.20; 95% CI 1.00- 1.43; p=0.046); only the ORBIT score was found to be statistically significant, but with weak discriminatory power at ROC curves (AUC 0.59; 95% CI 0.51–0.68; p=0.041). Conclusions In our cohort of elderly patients aged 75 or older, anticoagulated for NVAF, heart failure history was the only effective predictor of major bleeding risk during DOAC treatment. None of the bleeding risk scores used currently have demonstrated a good discriminatory power in our cohort. As predictive factors of bleedings in DOAC-treated patients may not be the same as those for VKA-treated patients and those in elderly may also be differ in younger people, it calls for more investigation on the topic.