The traditional independent component analysis is too difficult to solve the problems of underdetermined blind source separation(BSS) and statistically correlated sources separation existed in mechanical fault diagnosis.Assuming some sub-components of correlated machine vibration sources are independent,a novel blind source separation method based on subband extraction of ensemble empirical mode decomposition(EEMD) is proposed to solve the problem of single-channel statistically correlated mechanical signals separation. Firstly in this method,the single-channel signal is decomposed into a series of subband observed signals by ensemble empirical mode decomposition,then the number of source signals is estimated by singular value decomposition and Bayesian information criterion.Secondly,the new observed signals are reconstructed by the subband observed signals with high independence according to the mutual information criterion and the number of sources,the dimension of the new observed signal is increased.Finally,The source signals are estimated through the reconstructed observed signals by using whitening process and joint approximate diagonalization.The simulation and experiment testify the validity of the proposed method.