Atrial Fibrillation, a common arrhythmia accompanied by an increased morbidity and mortality remains difficult to treat either with medications or invasive procedures. Targeted destruction of atrial fibrillation triggers offers the best hope for permanent resolution of the arrhythmia. In this work, identification of atrial triggers is based on the analysis of complex endocardial recordings. Here, we propose a novel algorithm to detect the source of atrial fibrillation by classifying the signals originating from the four pulmonary veins in the left atrium.