Current　search　engines　are　not　very　effective　in　filtering　out　harmful　information　since　the　technology　they　use　for　filtering　is　often　based　on　traditional　text　classification　in　which　texts　are　often　classified　according　to　feature　words.　To　improve　the　effectiveness　of　filtering,　in　this　paper,　we　propose　a　new　filtering　scheme　in　which　we　combine　the　neural　network　and　ontology　categorization　techniques　to　improve　the　accuracy　of　classification.　We　show　that,　by　using　the　new　categorization　techniques,　the　accuracy　of　filtering　in　search　engines　can　be　greatly　enhanced　and　many　of　the　common　problems　can　also　be　resolved.　©　2010　IEEE.