The non?technical losses caused by electricity theft in the power system have always been a pressing issue for power grid companies to urgently address. With the deployment of a large number of smart meters in the power grid, the use of user?side data collected by the power metering automation system to accurately detect electricity theft has attracted widespread attention from researchers and power grid companies. Firstly, the basic classification of users' electricity stealing behavior, evaluation indicators and existing electricity theft detection data sets are introduced. Then, from the four aspects of grid state analysis, machine learning, game theory and hardware, the existing detection methods of electricity theft behavior are comprehensively sorted, analyzed and compared, and the basic ideas, advantages and disadvantages of each method are summarized. Finally, the current challenges in the field of electricity theft behavior detection are deeply analyzed, and a prospective outlook on the focus of future research work is provided.