Abstract：The traditional blind separation method of mechanical sources generally can not give a satisfactory separation performance under the unknown noise. Based on this deficiency, a new separation method of machine fault sources based on variational Bayesian independent component analysis (VbICA) is proposed. Compared with the traditional method, the proposed method has such unique characteristics, i.e. the unknown noise can not be regarded as an independent source. The denoising preprocessing is also neglected. The mechanical sources under the noisy environment can be directly separated. The simulation result shows that the proposed method is superior to the traditional method, the separation error is greatly reduced. The experiment results also verified the validity of the proposed method .