The invention relates to a smart power network neighborhood network malicious user detection method based on power stealing suspicion degree. The method comprises three phases: user suspicion degree evaluation, binary detection tree establishment and malicious user detection. In the user suspicion degree evaluation phase, user power stealing records are analyzed, and predicted values of power consumption of users are compared with reported values of the users, thereby analyzing power stealing possibility of the users. On the basis of the power stealing suspicion degree of the users, a binary detection tree taking the users as leaf nodes is established, and the binary detection tree is taken as a logic structure for assisting the search of the malicious users. In the malicious user detection phase, top-down and deep first search rules are employed. A sub-detector only carries out practical detection on left children of the binary detection tree. According to the detector provided by theinvention, the vast majority of logic nodes on the binary detection tree can be skipped, so the malicious users in a smart power network neighborhood area can be positioned rapidly and accurately.