Adaptive narrowband interference elimination technology and improved soft threshold denoising algorithm were used to process vibration signal of wind turbine, highlighting its fault characteristics. The vibration signal of wind turbines was mixed with periodic signal generated by the normal vibration source, fault vibration signal and noise signal. The periodic component of vibration signal could be eliminated reasonably by using the adaptive narrowband interference elimination algorithm. Then the background noise could be eliminated with improved soft threshold denoising technique. At last, the source signal with characterization of fault was highlighted.Tests show that the robustness and the generalization ability of the signal pretreatment algorithms are improved after considering adaptive characteristics. It lays solid foundation for further signal processing.