Aiming at the vibration signals fault feature of gearbox bearing was hard to extract, a method based on complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) and Mahalanobis distance was proposed. The vibration signal was decomposed by CEEMDAN to get several intrinsic mode functions (IMFs). Then, the noise and false mode functions were filtering by autocorrelation function and correlation coefficient, the components that characterized the signal were obtained. Finally, the energy entropy of each sensitive fault component was calculated, forming a state feature vector as a feature parameter; the working condition and fault type of the bearing were diagnosed by using the Mahalanobis distance method.The effectiveness of the proposed method is proved by analyzing the gearbox bearing signals measured under different working conditions and different fault degrees.