The early fault signal of hydraulic pump is characterized by non-stationary,strong background noise and weak fault characteristics,so it is difficult to extract fault characteristics effectively.Accordingly,a fault feature extraction method of gear pump was proposed based on autocorrelation analysis and maximum correlation kurtosis deconvolution algorithm.The collected signal was denoised by using the MCKD algorithm to enhance the original impact component of the signal and improve the signal to noise ratio.A feature information extraction method based on kurtosis (or absolute or square kurtosis value) was proposed to measure the non-Gaussian degree of mechanical signals and characterize the operating state information of mechanical equipment.The experimental results show that the proposed method is effective in extracting characteristic information of hydraulic pump fault signals.