Gastric　cancer　is　one　of　the　commonest　malignant　tumors　and　is　one　of　leading　causes　of　cancer　death　in　the　world.　Using　gene　expression　data　to　discriminate　tumor　from　the　normal　ones　is　a　powerful　method　and　will　have　a　strong　impact　on　disease　treatment　and　diagnosis.　In　the　paper,　we　present　a　hybrid　feature　selection　method　and　apply　it　to　a　29　gastric　cancer　dataset.　The　hybrid　method　contains　two　steps.　First,　we　rank　the　features　using　Bhattacharyya　distance,　then　we　delete　redundancy　based　an　adaptive　sequential　floating　forward　selection　(adaptive　SFFS).　In　the　experiment,　we　employ　the　support　vector　machine　(SVM)　to　recognize　the　gene　data　either　normal　or　tumor.　10　related　genes　have　been　selected.　The　accuracy　of　classification　shows　the　importance　of　the　10　genes　to　the　gastric　cancer,　and　certainly　shows　the　usefulness　and　effectiveness　of　our　hybrid　method.　©　2008　Binary　Information　Press.