Abstract Anticoagulation is recommended in Atrial Fibrillation (AF) patients for stroke prevention, and the bleeding risk associated suggests the need for a bleeding risk stratification. HAS-BLED score includes ‘labile INR’ referred to quality of anticoagulation. However, in naïve patients this item is not available. In addition, stroke and bleeding risk prediction scores shared several risk factors. Aims of our study were: 1) to evaluate if the HAS-BLED score in its refined form excluding ‘labile INR’ (HAS-BED) is still associated with bleeding risk; and 2) to evaluate the predictive ability for bleeding of both stroke and bleeding prediction models. We followed an inception cohort of 4579 AF patients enrolled in the START Register (NCT02219984). Major bleeds were recorded. During follow up (7014 pt-yrs) 115 patients experienced a major bleeding (rate 1.6 x100 pt-yrs). Patients at high risk were better identified by HAS-BLED and HAS-BED scores with respect to CHADS2 and CHA2 DS2 VASc. HAS-BLED has a slightly higher c value in comparison to CHADS2 and CHA2 DS2 VASc. However, among naïve patients the predictive ability for hemorrhage of HAS-BED score is overlapping with CHADS2 and CHA2 DS2 VASc. In low stroke risk patients (CHA2 DS2 VASc=0-1), only 6 patients are at high bleeding risk and none of them experienced major bleeding. In conclusion, in our prospective cohort of AF patients we found that HAS-BLED and HAS-BED scores identify patients at high bleeding risk. However, the predictive value for major bleeding of HAS-BED used in naïve patients is similar to CHADS2 or CHA2 DS2 -VASc, suggesting that stroke stratification scores could be sufficient for tailoring treatment.