Abstract: Aiming at the gear fault vibration signal often with lots of noise and fault feature of gear is weak, a gear fault diagnosis method based on maximum correlated kurtosis deconvolution (MCKD) and multiscale fuzzy entropy of improved Hilbert-Huang trasform is proposed. Firatly, the MCKD technique is used to eliminate the noise in the signal, to improve the signal noise ratio of the signal. Then, the signal is decomposed by complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN), obtain a series of intrinsic modal functions (IMFs) in different scales, and select the fault-sensitive modal component by the correlation coefficient-energy false IMF evaluation method. Finally, calculating the fuzzy entropy of sensitive IMF components, the obtained multiscale fuzzy entropy of the original signal is used as a state feature parameter input into the least squares support vector machine (LS-SVM) to diagnose the fault type of the gear. The results of a gear fault signals indicate that the proposed method can effectively diagnosis gear fault.