A fault classification method based on Sub band decomposition independent component analysis (SDICA) is presented for an engineering single sensor signal. The intrinsic mode functions obtained by the empirical mode decomposition(EMD) method are used as sub band signal, and are processed by independent component analysis (ICA) theory, the residual mutual information is extracted in the process of ICA separation, the approximate entropy is abstracted in the estimated sub band signal acquired by the SDICA, the characteristic parameters composed by residual mutual information and approximate entropy are used as input of the generalized regression neural network (GRNN) to realize the fault classification. The SDICA import the ICA theory in the single sensor signal actual fault classification, and the engineering single channel bearing signal achieve completely correct classification of three fault type, this method proves the residual mutual information and estimated subband approximate entropy can be importent classification character.