A method of fault diagnosis of gearbox based on the wavelet decomposition and least square support vector machine（LS-SVM）is proposed according to characteristics of vibration signal for gearbox.The reconstructed signal at the corresponding frequency ranges and the energy of each decomposed node are obtained by wavelet decomposition of gearbox vibration signal.The eigenvector composed by energy of each node is regarded as the eigenvector of diagnosis models and is put into the LS-SVM multi-classifier to recognize failures.The result of the recognition indicates that the common fault of wind turbine gearbox can be exactly identified by this method.