Aiming at the problem that the early fault characteristics of rolling bearing vibration signals with nonlinear and strong background noise are weak and difficult to identify, a rolling bearing early weak fault diagnosis method combining improved empirical wavelet transform （EWT）noise reduction with fast spectral correlation was proposed. In view of the problem that the EWT frequency band division method was greatly affected by noise and the division was unreasonable, the division method of maximum envelope processing was proposed; the improved EWT was used to perform adaptive decomposition of the signal to obtain intrinsic mode functions, and the kurtosis criterion was used to reconstruct each IMF to obtain the denoised signal; to enhance the period component of the early fault signal, the noise-reduced signal was analyzed by using fast spectral correlation, and the square enhanced envelope spectrum was obtained; the components with prominent amplitude in the squared envelope spectrum and the fault frequency were compared and analyzed to realize early fault diagnosis. The results show that compared with the fast spectrum analysis, the improved EWT noise reduction combined with the fast spectrum kurtosis diagram, by using the proposed method, the early fault characteristic frequency can be effectively enhanced, and the accurate diagnosis of the early fault can be realized.