The　finite　element　model　need　to　be　updated　according　to　actual　dynamic　responses　in　structural　damage　detection　and　health　monitoring.　The　model　updating　method　based　on　artificial　neural　network　(ANN)　is　improved　in　this　study.　Because　too　much　samples　are　demanded　for　training　ANN,　the　uniform　design　method　is　used　to　produce　samples　to　reduce　the　amount　of　samples　in　the　process　of　structural　damage　simulation,　thus　the　computation　efficiency　is　enhanced.　The　genetic　algorithm　is　used　to　optimize　the　initial　weight　of　back　propagation　network　and　as　a　result　the　operation　velocity　is　enhanced.　A　stepwise　model　updating　method　based　on　substructures　and　ANN　is　presented.　A　structure　is　divided　into　multilayer　substructures,　and　the　updating　is　taken　step　by　step　according　to　appropriate　damage　factors.　A　dome　structure　is　selected　as　an　example　and　two　damage　factors-the　frequencies　and　the　wavelet　packet　decomposition　(WPD)　energy　spectrums　are　investigated.　The　results　show　that　both　factors　can　give　accurate　results　by　using　this　method　but　the　latter　is　much　more　practicable.