This　paper　proposes　an　intelligent　damage　detection　strategy　in　terms　of　measured　acceleration　responses　in　practical　structures,　which　is　applicable　to　damage　location　and　extent　assessment.　Firstly,　genetic　algorithm　(GA)　and　the　Newmark　constant　average　acceleration　method　are　employed　to　identify　the　undamaged　stiffness　with　the　acceleration　responses　in　healthy　state;　Secondly,　acceleration　responses　in　damage　or　practical　states　are　processed　by　the　above　same　methods　so　as　to　identify　the　current　stiffness;　Thirdly,　after　the　damage　threshold　is　determined　by　the　hypothesis　testing,　damage　localization　and　extent　detection　are　performed　by　comparing　the　stiffness　difference　between　the　undamaged　and　current　stiffness.　Localization　results　through　the　validation　of　a　numerical　simulation　and　a　laboratory　experiment　agree　well　with　their　practical　cases.　This　damage　detection　strategy　not　only　reduces　the　complexity　of　the　detection　progress,　but　also　avoids　the　occurrence　of　the　time-frequency　domain　transform　error　owing　to　the　acceleration　responses　in　the　time　domain　being　constructed　the　fitness　function　directly.　Additionally,　the　maximum　relative　error　is　3.18%　for　damage　extent　prediction.　This　implies　that　the　intelligent　method　is　effective　and　applicable　to　damage　detection;　furthermore,　it　has　good　noise　tolerance　and　robustness.