Feature selection and parameters optimization of the fault classifier can enhance the fault diagnosis accuracy. Using the interdependent relationship between the feature selection and classifier parameter, a method of joint optimization of feature selection and classifier parameters is proposed to improve the diagnosis accuracy. By using the method we adopt the support vector machine (SVM) as a fault classifier, take into account of the radius-margin bounds for the accuracy evaluation of SVM classifier, and applies genetic algorithm (GA) to solve the joint optimization problem. In the gear fault diagnosis experiment, the joint optimization method guarantees better diagnosis accuracy and the optimization process has a higher rate than the single optimization of features or SVM parameters. So the joint optimization of fault features and classifier can fast achieve the better diagnosis accuracy in fault diagnosis.