Gene　expression　profiles　are　becoming　a　powerful　tool　for　clinical　diagnosis,　as　they　have　the　potential　to　discover　gene　expression　patterns　that　are　characteristic　for　a　particular　disease.　This　has　received　most　attention　in　tumor　classification.　In　this　paper　we　attempt　to　introduce　a　method　combined　neural　networks　with　two　feature　selection　mechanisms　for　tumor　classification.　Also　we　proposed　a　voting　weight　method　to　combine　the　classification　results　of　two　individual　neural　networks.　Then　we　validate　our　method　on　two　publicly　available　datasets.　Compared　with　other　current　methods,　our　method　greatly　improves　the　accuracy　and　robustness　of　such　classification.　We　hopefully　expect　that　the　biomarker　genes　analyzed　by　the　method　would　give　more　instruction　in　biological　experiments　and　clinical　diagnosis　reference.　©　2008　IEEE.