It is a challenging emission for variational mode decomposition to extract effective fault features since fault information in signals is weak and significantly affected by environmental noise. To address the described difficulties, a parameter-adaptive variational mode decomposition is proposed to extract fault features from bearings signals through solving the impact of artificially set parameters in the traditional method. Firstly, the sparrow search algorithm is employed to calculate the fitness function built on the mean envelope entropy for adaptively searching the number of modal decompositions and the penalty factor in variational mode decomposition when operating conditions are different. Secondly, both paraments obtained by the sparrow search algorithm help variational mode decomposition to decompose raw signal for obtaining intrinsic mode function components; and then those components are analyzed by the kurtosis criterion to reconstruct signal. Finally, envelope analysis is performed for the reconstructed signal to reveal a clear relationship between fault characteristic frequencies and harmonics. The proposed method is applied to machinery to demonstrate its effectiveness; the experiment results show that it can effectively reduce the impact of noise and extract fault features for bearing fault diagnosis.