Informative　gene　selection　is　of　great　importance　in　the　analysis　of　microarray　expression　data　because　of　its　huge　dimensionality　and　relatively　small　samples,　and　also　provides　a　systemic　and　promising　way　to　reveal　the　gene　expression　patterns　of　tumors　with　large　scale　gene　expression　profiles.　The　Multi-Class　tumor　gene　expression　profile　dataset　was　analyzed,　which　contains　218　tumor　samples　spanning　14　common　tumor　types,　as　well　as　90　normal　tissue　samples.　A　small　subset　of　genes　for　distinguishing　tumor　from　normal　tissues　was　found.　First,　a　Relief-based　feature　selection　algorithm　is　applied　to　create　candidate　feature　subsets　and　the　one　with　the　best　classification　performance　is　selected　as　the　informative　gene　subset　for　classification.　Then,　a　sensitivity　analysis　method　based　on　the　classifier　of　support　vector　machine　with　RBF　kernel　is　employed　to　eliminate　the　redundant　genes.　As　a　result,　52　informative　genes　are　selected　as　markers　for　making　distinctions　between　different　tumor　tissues　and　their　normal　counterparts,　and　their　expressions　are　analyzed　to　explore　the　tumor　gene　expression　patterns.　Several　methods　for　informative　gene　selection　are　also　analyzed　and　compared　to　validate　the　feasibility　and　effectiveness　of　the　method　employed　in　this　work.