This study presents an application of support vector machines (S VMs) to the analysis of electroencephalograms (EEG) obtained from the scalp of patients with epilepsy implanted with the vagus nerve stimulator (VNS) used in VNS Therapy®. The purpose of this study is to devise a physiologic marker using scalp EEG for determining optimal VNS parameters. Scalp EEG recordings were obtained from six patients with history of intractable partial onset epilepsy treated with VNS as adjunctive therapy to medicines. Averaged scalp EEG samples were used as features for separation. SVM classification accuracy was used as a measure of EEG similarity to separate a time segment during the beginning of stimulation from all the successive non-overlapping time segments within a full VNS on/off cycle. This analysis was performed for all the automated VNS cycles occurring during approximately twenty-four hours of 25 channels of scalp EEG. The patient that resulted in the lowest degree of EEG pattern similarity had the highest VNS stimulation frequency and experienced a monthly seizure rate among the lowest of all six patients included in this study. The patient with the greatest degree of pattern similarity had the lowest VNS stimulation frequency, shortest VNS pulse width, and experienced the greatest monthly seizure rate of all six patients included in this study. It is possible that VNS exerts its therapeutic effect by mimicking a theorized seizure effect for which a seizure has been observed to "reset" the brain from an unfavorable preictal state to a more favorable interictal state. These encouraging results suggest that data mining tools may be able to extract EEG patterns which could be used as an electrographic marker of optimal VNS stimulation parameters.