As　the　large　number　of　feature　attributes　in　Case-based　reasoning　system　(CBR)　brings　a　huge　information　redundancy　which　reduces　the　retrieval　efficiency,　a　novel　reduction　method　based　on　Water-Filling　is　proposed　to　remove　those　unnecessary　attributes.　In　the　method,　the　importance　of　each　attribute　could　be　calculated　by　utilizing　the　ratio　of　the　standard　deviation　and　the　mean　value　of　each　attribute　data　as　evaluation　parameter,　and　the　impotance　result　of　each　attribute　is　then　used　to　guide　the　reduction　process.　The　experiments　on　glass　identification　showed　that　the　new　method　could　get　a　better　retrieval　accuracy　as　well　as　a　greater　efficiency　compared　with　the　methods　which　do　not　conduct　the　reduction　process.　©　2012　IEEE.