The present invention belongs to the field of fault diagnosis methods, and discloses a rolling bearing early fault diagnosis method. The method comprises the steps of: performing preprocessing of collected fault signals; extracting fault signal feature frequency after preprocessing; and comparing the extracted feature frequency and a known fault feature frequency, and identifying a fault type. Thepreprocessing comprises two aspects consisting of noise reduction and reduction of disturbance vibration components. In the feature extraction step, fault signals after preprocessing are subjected tosegment processing, the segment fault signals after preprocessing and fault-free bearing signals are subjected to correlation analysis, and signal segments with many fault signals are selected according to cross correlation coefficients to perform frequency feature extraction. The rolling bearing early fault diagnosis method can improve the accuracy and the effectiveness of rolling bearing earlyfault diagnosis, can provide a new way for solution of the rolling bearing early faults, and can be widely applied to complex mechanical system fault diagnosis in the fields such as chemical engineering, metallurgy, machining, aviation and the like.