In　order　to　make　full　use　of　the　redundant　and　complementary　information　and　thus　assess　the　structural　health　states　from　a　large　structural　health　monitoring　system,　the　principle　of　data　fusion　was　first　introduced　in　this　paper,　then　a　5-phase　novel　decision-level　data　fusion　damage　detection　approach　by　integrating　wavelet　analysis,　probabilistic　neural　network　(PNN)　and　data　fusion　developed　and　implemented.　Finally,　two　numerical　examples　validated　the　proposed　method,　the　effect　of　measurement　noise　on　identification　accuracy　was　investigated　as　well.　The　result　shows　that　the　proposed　method　is　feasible　and　effective　for　damage　identification.