This　paper　presents　an　experimental　study　on　using　an　improved　extended　Kalman　filter　(EKF)　to　identify　impact-induced　structural　damage.　By　introducing　the　optimization　of　estimated　residual　error　into　the　classical　EKF,　this　real-time　approach　demonstrates　an　excellent　capability　to　identify　the　abrupt　changes　of　structural　parameters　instantly　and　accurately.　The　optimization　procedure　is　activated　when　a　prescribed　threshold　is　exceeded.　A　shaking　table　test　of　a　three-story　steel　frame　subjected　to　abrupt　damage　induced　by　impact　load　was　conducted　to　validate　the　improved　EKF　approach.　The　results　clearly　reveal　its　improved　performance　and　good　anti-noise　ability　in　identifying　time-variant　structural　parameters.