To solve the problem of epistasis analysis in genome-wide, a filter method based on multiple criteria fusion was developed to remove the unrelated SNP loci. After that, ant colony algorithm was used to construct the SNP set with epistasis interaction. In the phase of constructing, a support vector machine was proposed to build the relationship between the SNP set and complex diseases. The experiment results show that, with multiple criteria evaluating each SNP and ant colony optimization, the prediction accuracy and running time have been improved, making it better than conventional methods.