Abstract: In order to achieve electricity theft detection, this paper proposes an anti-electricity theft alarming method based on non-intrusive load monitoring through the analysis of electricity consumption behavior of users. In this method, the commonly-used process is carried out, including the load event detection, feature extraction and meanshift clustering method, so as to obtain the load features, categories and other information of users. The dataset of load category comparison and the probability prediction model of electricity theft detection are built. Meanwhile, according to the electricity theft detection model, the prediction is performed by the information, including the load switching event, the length of load work and the energy consumption. The Bayes theory is then introduced to infer whether its electrical consumption behavior is normal or not. The experiments are carried out by using the real information from the smart meter. The results show that the proposed method can provide the basis and support for the electricity theft detection, which lays a foundation for the application of anti-electricity theft of a new generation smart meter.