The invention discloses a sample feature dimension reduction processing method for abnormal behavior detection of an electric power monitoring system. Feature indexes are selected from two aspects ofcategory correlation and feature redundancy. Firstly, historical data is labeled manually, and a training sample data set is constructed; whether each system behavior characteristic index in the sample data set generates correlation with the abnormal behavior or not is analyzed, and the correlation priority of each characteristic index on the abnormal behavior judgment influence is determined; Then, starting from the feature index with the highest correlation priority, redundancy between every two feature indexes is sequentially checked, the feature index which has small influence on system abnormal behavior judgment and high feature redundancy is removed, and finally the effect of feature dimension reduction processing is achieved. The method can be used for reducing the scale of trainingsamples by carrying out dimension reduction processing on the behavior characteristic index set of the power monitoring system, and provides better service for behavior recognition classification training of subsequent abnormal detection.