Aiming at the problem that the fault feature of hydraulic system is not clear and difficult to diagnosis,a feature extraction method of faults is proposed in this paper which combines EEMD with correlation analysis. First,a group of IMF components of hydraulic fault signal was obtained by EEMD. Screened the IMF components strongly correlated with fault signal as a group of faults’ candidate IMF component set. Then screened the IMF components little correlated with normal signal as another group of faults’ candidate IMF component set. The intersection of two candidate IMF component sets were the main fault features of the original fault signals of IMF components. The main fault characteristic frequency of fault signal could be got by the spectrum analysis. Then use the cross-power spectrum of fault IMF component and normal signal to prove the validity of the feature extraction. The fault feature IMF components and fault frequency of hydraulic system leakage fault of the hydraulic simulator are extracted in this paper.