Based　on　the　fact　that　the　number　of　elements　is　much　larger　than　that　of　the　nodes　in　a　grid　structure,　preliminary　localization　method　of　the　nodes　oriented　structural　damage　and　a　three-step　damage　identification　strategy　for　large　grid　structures　are　used　in　the　damage　identification　process　based　on　BP　networks.　The　vibration　parameters　of　a　double-layer　cylindrical　latticed　shell　model　are　measured　under　four　different　damage　cases.　The　variation　rate　of　the　measured　lower　frequencies　and　components　of　the　lower　mode　shapes　moasured　from　a　few　nodes　of　the　lattice　shell　model　are　used　as　the　input　parameters　of　the　BP　neural　network,　the　identification　of　different　damage　cases　are　carried　out.　The　results　indicate　that　the　method　used　in　this　paper　can　greatly　reduce　the　architecture　of　the　BP　networks,　and　thus　the　pattern　recognition　capability　of　the　BP　networks　can　be　remarkably　improved.　These　methods　can　be　used　for　the　damage　identification　of　large　complex　structures　with　significant　efficiency.