Voltage sag is one of the major power quality (PQ) problems, and has been the focus of PQ studies due to the impact on sensitive industrial loads and costs led by the damages and maintenance. Voltage sag source location is significant for the customers and suppliers to solve the issue between them, as well as for possible mitigation. Five main methods (the disturbance power and energy method, the slope of system trajectory method, the real current component method, the resistance sign-based method, and the distance relay method) are reviewed first. However, these methods used single criteria, and their effect is limited as the literature shows. This paper presents a pattern recognition way to locate the source of voltage sag. In the proposed method, features for pattern recognition are extracted first, based on these five methods. Then, the thought of source location by pattern classification is discussed with three steps in detail, and support vector machine (SVM) is applied in the case. The nonlinear binary classifier with optimal hyperplane is established to classify the sag source from upstream or downstream by SVM learning. To illustrate the effectiveness of the proposed method, a 110-kV distribution system is tested under simulation conditions, and records from PQ monitors installed in 35-kV substations are used. (C) 2013 American Society of Civil Engineers.