Gene　selection　based　on　gene　expression　profiles　is　of　great　importance　in　oncology　and　cancer　treatment　and　has　recently　received　a　great　deal　of　attention　in　the　field　of　bioinformatics.　A　new　method　of　gene　selection　utilizing　SVM　(Support　Vector　Machine)　methods　based　on　sensitivity　is　proposed　in　this　paper.　First,　sensitivity　of　a　discriminant　function　with　respect　to　a　component　of　the　input　vector　in　SVM　is　defined.　Then　this　method　is　formulated.　In　this　method,　after　training　and　testing　a　SVM　model　using　a　set　of　samples,　the　genes　are　weighted　by　their　sensitivity　value,　and　those　with　less　sensitivity　value　are　discarded.　This　process　is　repeatedly　executed　until　the　number　of　genes　decreases　to　zero.　The　set　of　selected　genes　is　the　one　that　has　the　best　classification　performance　and　least　number　of　members.　This　method　was　applied　to　gene　expression　profiles　of　cancerous　and　normal　colon　tissue.　A　comparative　experiment　showed　that　the　gene　set　selected　by　this　method　was　more　powerful　for　classification　than　those　reported　in　certain　previous　research.