The bearing is an important part of aircraft power system and mechanical structure, which has complex fault characteristics and has the crucial impact on flight safety. In order to improve the accuracy of bearing fault diagnosis, a bearing fault signal analysis method based on empirical wavelet transform and spectral negative entropy is proposed in this paper. The method firstly calculated the time-domain spectral negative entropy and frequency-domain spectral negative entropy of each component after decomposing the fault signal by empirical wavelet transform. In view of the fact that the average spectral negative entropy cannot adjust the impact and periodic weight coefficients adaptively, an adaptive average spectral negative entropy is proposed. Then using the kurtosis maximization as fitness function, grey wolf optimization algorithm is used to optimize the number of components and the adaptive average spectral negative entropy proportional coefficient to reconstruct the fault signal. The results show that:1) the reconstructed signal retains fault impact components of fault characteristics;2) after the fault signal reconstruction, the kurtosis and signal-to-noise ratio are significantly improved; 3) both time-domain spectral negative entropy and kurtosis can quantify the impact components of the signal, however kurtosis is more sensitively affected by noise.