Abstract: In distribution lines/stations, there is a causal relationship between the electricity consumption of access users and the line loss. The change of the normal users' electricity quantity has a limited influence on the line loss; while the users' power theft will make the influence of the electricity consumption on the line loss different from that of the normal users. Transfer entropy can measure the information transfer between variables, and it is an important index to evaluate causality. This paper proposed a method to identify electricity stealing users based on transfer entropy density clustering. Firstly, the users with strong causal correlation to line loss electric quantity in line/station areas were selected by using transmission entropy information directivity. Then, the transfer entropy model between electricity consumption and line loss quantity was constructed to calculate the transfer entropy value of electricity consumption of different time length to line loss quantity to measure its information transfer quantity. Combined with the density clustering algorithm, the users whose transfer entropy curve deviates from the normal user cluster were identified as those who steal electricity with strong causality to line loss. Finally, the effectiveness of the proposed method was proved based on the verified data of high-loss stations and high-loss long-distance distribution lines.