Harmonic sources identification is needed when the information about the positions and number of harmonic sources in the network is insufficient. To solve this problem, a harmonic sources identification method is proposed. Sensitivity index is used firstly for the optimal placement of harmonic measurement devices. Then a method based on genetic algorithm (GA) and support vector machine (SVM) is used to classify the buses as suspected and non-suspected ones. GA is used for parameters optimization for SVM models. At last, harmonic power flow is performed to examine the accuracy of the predicted locations. To evaluate its reliability and effectiveness, the proposed method is applied to the modified IEEE 13-bus system. The evaluation results are very promising. This work is supported by Science and Technology Program of State Grid Shandong Electric Power Company (No. 520625150002).