Gene　expression　profiles　of　gastric　cancer　and　counterpart　normal　tissues　were　analyzed　with　bioinformatics　and　machine　learning　methods　to　address　the　problem　of　discovery　of　biomarker　genes　and　model　of　the　tumor　molecular　diagnosis.　Firstly,　a　support　vector　machine　(SVM)　was　employed　to　find　the　feature　gene　subset　with　best　classification　performance　for　distinguishing　cancerous　tissues　and　the　counterparts.　And　then,　using　genetic　algorithm　filter　feature　genes　based　on　the　classification　rate　of　SVM.　The　decision　tree　was　employed　to　extract　rule　subsets　from　the　feature　genes.　These　rules　were　tested　by　the　test　dataset　with　cross　validation　method,　and　the　diagnosis　model　was　constructed　with　these　rules.　The　results　indicate　the　gastric　biomarkers　and　the　diagnosis　model　would　give　more　instruction　in　biological　experiments　and　clinical　diagnosis　reference　model.