Abstract: Aiming at the problem of parameter setting in the process of extended singular value decomposition packet, an improved method was proposed to select parameters adaptively. The spectrum trend was estimated by calculating the key function of the signal, to realize the adaptive setting of two important parameters in the original method: decomposition accuracy and decomposition layer. Besides, the time-domain negative entropy index was introduced to reduce the noise in the decomposed components, and to improve the signal-to-noise ratio of the components, which lays the foundation for the later feature extraction and fault diagnosis. The simulation and experimental results show that the proposed method can effectively extract the fault features of vibration signal and realize the fault diagnosis of rolling bearings.