Atrial fibrillation (AF) patients are widely heterogeneous in terms of ischaemic stroke risk, and several risk stratification schemes have been developed. We performed a prospective study on 662 AF patients on long-term oral anticoagulant therapy (OAT), evaluating the agreement among the different schemes and their correlation with adverse events recorded during follow-up. Patients at low risk were similarly distributed among the different models. Instead, patients classed at moderate risk were 49.2% by CHADS(2) score, 27.6% by NICE and 2.3% by ACCP. As a consequence patients classed at high risk were 46.1% by CHADS(2), 69.8% by NICE and 95.3% by ACCP. CHADS(2 )and NICE scores were associated to the best predictive accuracy. A separate analysis was performed for patients on treatment for secondary prevention, and we observed that they were included in high risk groups by all models, except for 14 patients (6.3%) classed at moderate risk by CHADS(2) even though these patients are at very high risk and the use of aspirin could be unsafe for them. During follow-up 32 major bleeding (1.35 per 100 patient/years) and 39 thrombotic events (1.64 per 100 patient/years) were observed. Among patients on OAT for secondary prevention, both bleeding and thrombotic events mostly occurred in high-risk patients. Even if the absolute rate of adverse events is low, this finding seems to confirm the high stroke risk of this group of patients. For patients on secondary prevention there is no need for further stratification and warfarin should be the treatment of choice.