Abstract: The traditional mechanical fault source separation method is restricted to nongaussian, stationary and mutually independent source signals, and the number of observations is assumed to be more than the number of sources.These deficiencies may cause many problems in the application of fault diagnosis, because fault signals generally do not meet these conditions.Based on these deficiencies, an underdetermined blind separation method of machine faults based on empirical mode decomposition(EMD) was proposed.The proposed method was that the mixture signals were decomposed into a series of intrinsic mode functions(IMF) by the EMD method.These IMFs and original observations were composed into new observations, which was blindly separated.The simulation and experiment results show that the proposed method is very effective.