Gearbox signals always tend to be drowned in the noise at a complex environment, resulting problem in feature vector extraction/which make difficulty for feature vector extraction. In order to effectively diagnose gearbox fault, a method for gearbox is proposed based on maximum correlated kurtosis deconvolution (MCKD) ensemble empirical mode decomposition (EEMD) approximate entropy and twin support vector machine (TWSVM). Firstly, the noise signal is filtered by MCKD method, then EEMD method is used to decompose the signal of gearbox. To get signal eigenvectors, the intrinsic mode functions (IMF) obtained from the decomposition is used to calculate approximate entropy. Finally, signal eigenvectors are input to the TWSVM classifier for fault identification. Simulation results show that the MCKD-EEMD method can be used to extract raw signal effectively, and the TWSVM classifier performs a better classification.