element model in conjunction with a recently proposed damage detection technique, referred to as the adaptive quadratic sum-square error with unknown inputs (AQSSE-UI). The identification process is divided into two steps. In the first step, static condensation technique is used to reduce the order of the equations of motion of the finite-element model. In the second step, the adaptive quadratic sum-square error with unknown inputs (AQSSE-UI) is used for the on-line system identification and damage detection of the reduced order system. The proposed approach is capable of identifying time-varying parameters of linear or nonlinear hysteresis structures. The capability of the proposed damage detection technique is demonstrated by shake table test data using large-scale structures. A 1/3-scaled 6-story steel frame, a 1/3-scaled 2-story RC frame and a 1/2-scaled one-story two-bay RC frame have been tested experimentally on the shake table at NCREE (The National Center for Research on Earthquake Engineering), Taiwan. For the 1/3-scaled 6-story steel frame structure, the damages of the joints were simulated by loosening the connection bolts. The 1/3-scaled 2-story RC frame was subject to a sequence of earthquake excitations back to back. Both RC frames are modeled by a series of finite elements and plastic hinges following the generalized Bouc-Wen model. Experimental results demonstrate that the proposed damage detection technique is quite accurate and effective for the tracking of: (i) the stiffness degradation of linear structures, and (ii) the non-linear hysteretic parameters with stiffness and strength degradations.