The invention discloses an electricity stealing detection method and system based on an SOM neural network and K-means clustering. The electricity stealing detection method comprises the steps: randomly extracting data from a user load curve to serve as a training sample, carrying out normalization processing on the training sample, and obtaining a processing sample; clustering the processing sample based on an SOM neural network, and obtaining the number of clusters and an initial clustering center of the processing sample; taking the clustering number and the initial clustering center as theinitial value of K-means clustering; based on the K-means clustering, clustering the processing sample to obtain a user load feature curve; calculating the Euclidean distance between the to-be-detected user load and the user load feature curve thereof, and obtaining the Euclidean distance between the to-be-detected user load and the user load feature curve thereof; and when the Euclidean distanceis greater than a preset threshold value, judging the user as a suspected electricity stealing user.