The　belief　rules　are　stored　out　of　order　in　the　extended　belief　rule　base　(EBRB),　which　will　weaken　its　reasoning　performance　in　that　all　rules　are　visited　when　calculating　each　rule＇s　activation　weight.　This　paper　focuses　on　reducing　the　number　of　rules　which　are　visited　in　the　calculation　of　each　rule＇s　activation　weight.　A　new　rule　activation　method　based　on　VP-tree　and　MVP-tree　is　proposed　to　build　index　structure　to　store　rules.　The　proposed　rule　activation　method　is　based　on　rule　similarity　query,　where　only　partial　rules　will　be　retrieved　and　visited　while　calculating　each　rule＇s　activation　weight.　Note　that,　the　performance　of　EBRB　systems　based　on　tree　index　is　affected　greatly　by　the　value　of　query　threshold.　However,　sometimes　it　is　difficult　to　determine　the　value　of　query　threshold,　so　this　paper　also　proposes　an　approach　based　on　the　k-means　clustering　algorithm　to　choose　the　appropriate　query　threshold.　Some　case　studies　show　how　the　use　of　the　proposed　optimization　method　enhances　the　reasoning　performance　of　EBRB　systems.　The　proposed　method　has　been　validated　to　be　advantageous　to　visit　partial　suitable　rules　instead　of　all　rules.　Beside　the　work　performed　in　the　EBRB,　the　proposed　method　alone　can　also　be　used　in　different　application　areas.