BACKGROUND Echocardiographic parameters for predicting cardioversion outcome in patients with non-valvular atrial fibrillation are not accurately defined. OBJECTIVE To evaluate the role of left atrial appendage flow velocity detected by transoesophageal echocardiography for prediction of cardioversion outcome in patients with non-valvular atrial fibrillation enrolled in a prospective. multicentre, international study. METHODS Four hundred and eight patients (257 males, mean age: 66 +/- 10 years) with non-valvular atrial fibrillation lasting more than 48 h but less than 1 year underwent transthoracic echocardiography and transoesophageal echocardiography before either electrical (n=324) or pharmacological (n=84) cardioversion. RESULTS Cardioversion was successful in restoring sinus rhythm in 328 (80%) and unsuccessful in 80 patients (20%). Mean left atrial appendage peak emptying flow velocity was significantly higher in patients with successful than in those with unsuccessful cardioversion (32.4 +/- 17.7 vs 23.5 +/- 13.6 cm x s(-1); P<0.0001). At multivariate logistic regression analysis, three parameters proved to be independent predictors of cardioversion success: the atrial fibrillation duration <2 weeks (P=0.011, OR=4.9, CI 95%=1.9-12.7), the mean left atrial appendage flow velocity >31 cm x s(-1) (P=0.0013, OR=2.8, CI 95%=1.5-5.4) and the left atrial diameter <47 mm (P=0.093, OR=2.0, CI 95%=1.2-3.4). These independent predictors of cardioversion success outperformed other univariate predictors such as left ventricular end-diastolic diameter <58 mm, ejection fraction >56% and the absence of left atrial spontaneous echo contrast. CONCLUSIONS In patients with non-valvular atrial fibrillation, measurement of the left atrial appendage flow velocity profile by transoesophageal echocardiography before cardioversion provides valuable information for prediction of cardioversion outcome.