Abstract： A novel approach to fault diagnosis of rolling bearing using singular value decomposition (SVD) and multiple deep belief network (DBN) classifiers was proposed. According to this approach, vibration signals of rolling bearing under different conditions were reconstructed in the phase space and characteristic matrixes were obtained. Then, the characteristic matrixes were decomposed by SVD to get the singular values. After that, all the singular values were used to form a characteristic vector. Finally, a multiple DBN classifiers model was developed to identify the faults of rolling bearing. To confirm the superiority of the proposed approach, it was compared with DBN, BP neural network and SVM. The experimental results indicates that the proposed approach has a better performance in accuracy and efficiency to identify the fault patterns and severity of rolling bearing.