DNA microarray data often contain tens of thousands of genes, where have a lot of irrelevant and redundant genes, these genes may seriously affect the accuracy and efficiency of classification. In order to solve this problem, this paper proposes a feature gene selection method based on improved harmony search algorithm. Firstly, genes are ranked using Relief F algorithm and preselected genes subset is obtained according to ranked-top genes .Then using improved harmony search algorithm to select feature genes from above genes subset. Finally we implement simulation experiments on three public microarray data sets. The results show that the proposed algorithm can achieve very high accuracy in the feature genes less, and is a effective and efficient feature genes selection algorithm. Especially, selected feature genes can help to understand microarray data.