Based　on　wavelet　packet　decomposition,　rough　set　theory　for　feature　reduction　and　the　strong　classifying　power　of　support　vector　machine　(SVM),　a　new　approach　of　structural　damage　diagnosis　for　Spatial　Structures　is　presented.　The　purpose　is　to　better　the　existing　damage　diagnosis　approaches,　reduce　the　cost　of　structural　health　monitoring　and　improve　the　accuracy　and　speed.　The　frame　of　the　spatial　structural　Health　Monitoring　system　is　established.　The　damage　diagnosis　used　by　this　approach　for　a　single-layer　spherical　lattice　shell　is　simulated　by　finite　element　analysis.　The　results　show　that　the　feature　vectors　gained　by　wavelet　packet　decomposition　could　reflect　the　element　damage　sensitively,　and　the　support　vector　machine　after　being　instructed　could　identify　the　location　and　degree　of　the　elements　precisely.　The　damage　diagnosis　based　on　rough　set　reduction　has　nearly　the　same　results　while　the　computing　speed　is　enhanced.　This　approach　has　some　advantages,　such　as　engineering　orientation,　low　cost　and　convenience.