Support Vector Machines（SVM） is a new machine learning method based on statistics learning theory which has good classification ability even if the training samples are very few.The technology of SVM is applied to gear fault diagnosis.Using of local

global kernel function for samples training classification

then selects their respective weight appropriately by using the previous classification results to construct mixed kernel function.Experiments and data analysis prove that the classification results using mixed kernel function are better than that using local or global kernel function alone.