The invention discloses a reduction outlier algorithm-based electricity stealing identification method. The method comprises the following steps of removing redundant attributes by utilizing an attribute reduction technology; obtaining a same or similar outlier data set in a smaller attribute sub-space; determining an outlier set through calculating an adjacency-based outlier weight; and judging user history load data corresponding to all the data points of the outlier set as electricity stealing data according to relation between a history data set and an electricity utilization data set, and outputting the electricity stealing data. The invention furthermore discloses a device of the method. According to the method and the device, the electricity stealing users can be identified rapidly, effectively and correctly, so that a new way is provided for power grid workers to rapidly and effectively carry out anti-electricity stealing analysis by utilizing data of an electricity utilization information acquisition system.