The invention relates to an anti-electric theft method based on clustering discrete point detection, in particular to a method for collecting electric data usage in a user electric power system. Cleaning data, data reduction and data transformation processing, namely data preprocessing, are performed on the collected electric power data. Secondly, using a density-based k-mean algorithm, the data detection model is built and the pre-processed data is input. Finally, the power consumption data after clustering is further tripped in, and the suspected users are judged according to the tripped-inresults. The invention adopts the density-based k-mean method through collecting the usage situation of the power consumption data in the power consumption system of the user. The k-mean algorithm asthe core algorithm is used to establish a data judgment model, which can accurately screen out suspected users of electricity theft, narrowing the scope of the investigation, and saving the company'sinvestment in electricity theft.