Abstract: Abstract: Support vector machine(SVM) has been widely used in data mining as a classification algorithm. However, there is no effective standard in choosing its kernel function and setting the related parameters. In order to reach both stronger learning ability and generalization ability, this paper poposes a hybrid kernel function, in which the parameters of SVM are optimized by means of PSO(particle swarm optimization). The comparison is made on a classical classification problem, which is shown from the results that the proposed algorithm has better performance than the SVM based on single kernel function.