Support vector machine (SVM) is a machine learning method based on statistical learning theory proposed by Vapnik in the mid-1990s. Different from the traditional method based on the empirical risk minimization criterion, SVM considers both the empirical risk and the structural risk, which can better complete the optimal classification decision in many cases. Based on the outlier theory, this paper analyzes the electrical data of a large number of users systematically from the macro perspective, and obtains the suspected degree of power stealing, which provides an effective guide for power stealing inspection. However, limited by the existing system collection data items and collection frequency, the collected data is difficult to fully present the user’s power consumption information. In order to fulfill the needs, this paper investigated several detection methods which detect abnormal usage of electricity using SVM. By collecting and analyzing users’ electricity consumption data for a sufficient period of time, the unusual consumption pattern can be effectively identified from mass original data.