Electricity thieves show new characteristics such as stealing only part of electricity and intermittent occurrence, becoming harder to detect. This paper proposes a remote detection method for electricity theft based on dynamic correlation factor and optimized shift-splitting iteration method. Firstly, through mathematical analysis and verification of the electricity theft mechanism, we found an intermittent and nonlinear positive correlation between the line loss rate and electricity thieves’ electricity consumption proportion. Based on this characteristic, the dynamic correlation factor is used to identify suspected electricity thieves. Secondly, a low-dimensional model is established according to the generalized conservation of electricity, and the optimized shift-splitting iteration method is used to solve the ill-conditioned model, realizing accurate identification of electricity thieves. Through two-step identification and calculation, the algorithm addresses the problem that the intermittency of electricity theft, disturbance under complex scenes and the error of parameters estimation can lead to decrease of accuracy and increase of missed detection rate. Case studies in distribution areas of a province in China show that the proposed method is reliable, and has higher accuracy (over 90%) and lower missed detection rate (below 10%) than previous algorithms, especially in large distribution areas with complex scenes.