Abstract: This paper describes a method for extracting fault features from vibration signals based on fast independent component analysis (FastICA) and ensemble empirical mode decomposition (EEMD) to overcome the problems of mutual influence between vibration signals of a hydro-generator unit and noise disturbance to them. First, a vibration signal is decomposed into independent components by FastICA; then, each component is processed by EEMD and its effective intrinsic mode functions (IMFs) are selected by calculating the normalized energy and normalized correlation coefficients of all the IMFs; finally, these effective IMFs are used to extract its fault features. The results of simulation and application show that in comparison with other methods, this method can effectively suppress noise interference and it is more comprehensive and accurate in extraction of vibration feature signals for hydro-generator units, fully meeting practical engineering demands.