The support vector machine（SVM） can avoid the overlearning phenomenon in the case of small training samples,so that the generalization ability can be maximized. The problem that the SVM parameters cannot be selected adaptively is studied. By using the global search characteristic of the genetic algorithm,the parameters of the support vector machine are optimized and the optimal parameters of the support vector machine are obtained. This method is used to study the acoustic emission detection of gear fault,and the experimental system is composed of PCI-2 acoustic emission system and rotary machinery vibration fault simulation test bed. The accuracy of gear fault classification is 10% higher than that before optimization,which is very important for gear fault diagnosis.