Alternative splicing is an important mechanism to increase the transcript diversity. Previous studies which based on expressed sequence tags (ESTs) and microarray analysis have identified many alternative splicing events, and have shown that alternative 5′/3′ splice sites occur frequently in the human genome. Because of the experimental limitations, computational approaches for identification of alternative 5′/3′ splice sites are needed. In this paper, the support vector machine method which combines with position weight matrix and increment of diversity is proposed. And based on the mechanism of splice site competition, more than 71% of alternative splice sites and constitutive splice sites in the human genome can be correctly classified.