In the early fault period of rolling bearing, the characteristic signal is weak, and fault identification is very difficult. In order to solve this problem, a diagnostic method based on improved variational mode decomposition and singular value difference spectrum was proposed. The original signal was decomposed into several intrinsic mode function components after being processed by improving the variational mode decomposition method. Then, the best component was selected by the sparsity index of the envelope spectrum. The Hankel matrix was constructed through the best component, and singular value decomposition was operated. After acquiring the singular value difference spectrum, the signal was reconstructed through the maximum catastrophe point, and the fault type of bearing was judged by analyzing the envelope spectrum of the signal. The method was used to analyze simulated and measured signals of the fault bearing based on improved variational mode decomposition and singular value difference spectrum, and the weak characteristic information was extracted successfully. The results show that the proposed method can judge the incipient fault of rolling bearing effectively and have a certain reliability and application value.