For purpose of more effective extraction of feature of hydraulic pump, an approach based on Modified High Frequency Harmonic AddedEmpirical Mode Decomposition (HFHAEMD) and Two Correlation coefficient Criteria was proposed. Firstly, high frequency harmonic of higher amplitude was added to vibration signals, which was decomposed by EMD, and a group of Intrinsic Mode Function (IMF) was obtained. Secondly, every IMFs Correlation coefficient with high frequency harmonic and vibration signals was calculated, and some of them which were up to standard were selected to reform vibration signals. Thirdly, reformed vibration signals were conducted to demodulation, and its spectrum envelope was obtained. Finally, each frequency band's total energy percentage of all band of the spectrum envelope was constructed as feature extraction, which was put into support vector machine (SVM) to pattern recognition of the pump. Experimental results indicat that Modified HFHAEMD restrain mode aliasing efficiently than HFHAEMD. Meanwhile, two correlation coefficient criteria wipe out IMF which is unconcerned with fault and keep IMF which is concerned with fault to make a solid base to feature extraction of the hydraulic pump.