Abstract: A seismic Disaster causes serious lose of people’s life and properties. Building assessment is then one of the important works to rebuild the effected area. The globe property identification based on vibration data could be a critical supplement to the assessment technical only based on local damage identification. This paper studies the performance of the prediction error method (PEM)-based second order structural identification algorithms through a 3-storey model structure vibration test. By applying PEM method to identify the ARMAX/ARMA model derived from state space model, this identification method can identify second order model parameters such as mass, stiffness, damping ratios directly from measured vibration data. This system identification method can be used for investigating the structural properties and has potentials to be applied as a damage detection algorithm on structural health monitoring systems. To study the performance of this method, a 3-storey model structure is built. The identification results from measurements of vibration tests on this model structure including top story excitation tests and ground motion tests show that the PEM based second order structural system identification technique is capable of accurately identifying the second order parameters and locating and quantifying damages reasonably well in the context of real vibration test data.