A method to extract the energy features of frequency bands based on EMD-AR spectrum estimation was proposed by combining the empirical mode decomposition (EMD) and AR power spectrum parameter estimation model.The EMD-AR spectrum curve was obtained by empirical mode decomposition of the vibration signal and AR spectrum estimation of the top groups of intrinsic mode functions (IMFs) that accounted for the larger energy after decomposition;then,the EMD-AR spectrum curve was segmented according to the frequency band and the energy of each segment was summed up and normalized as the feature value.By building a gearbox fault test bench,vibration signals were collected for EMD-AR spectrum frequency band energy feature extraction,and the obtained feature values were input into SVM for training and diagnosis.The results show that the fault diagnosis model obtained by this feature extraction method combined with SVM has a high recognition accuracy and can effectively and accurately perform diagnosis and recognition.