A series of inverted Event-Related Potential (ERP) signals were analyzed with pattern recognition methods using new morphological features and powerful classifiers, in an attempt to develop a computer-aided discrimination system of OCD patients from controls. Eighteen OCD patients and twenty controls were examined. All subjects were evaluated by a computerized version of the digit span subtest of the Wechsler Adult Intelligence Scale. EEGs were recorded from 15 scalp leads, and they were inverted to intracranial current sources using the BET-ART method. 60 source locations were positioned on a spherical shell corresponding to the outer layers of the cortex. From the P600 component of each signal nineteen waveformfeatures were calculated. The Probabilistic Neural Network (PNN) classifier was developed, and it was fed with feature combinations from all source locations. Highest overall accuracy (84.2%) was found at the right frontal brain region indicating that OCD patients may present deficits related to working memory mechanisms corresponding to that region.