Abstract: We describe a novel approach for gene selection and cancer classification of microarray data,which combines Support Vector Machines （SVM）,Correlation-based Feature Selection（CFS） and Genetic Algorithms（GA）.First,the Bhattacharyya distance of each gene is used as the criterion for filtering the irrelevant genes for classification.Then GA combined with CFS is adopted to find informative gene subsets.Finally,using archive records of these subsets,the 50 most frequently selected genes are defined as a candidate subset through which the GA is used to evolve gene subsets whose fitness is evaluated by a SVM classifier.Our method is assessed on the colon dataset and is able to select small subsets and still improve classification accuracy.