Abstract: In order to evaluate the ability of a building to maintain and to recover its original functionality after an earthquake, the seismic resilience assessment based on nonlinear time-history analysis has undergone a rapid development in recent years. Multi-degree-of-freedom (MDOF) models, featured high computational efficiency, have demonstrated a wide application potential in large-scale computational tasks such as regional seismic damage simulation and data-driven assessments. Although existing research generally acknowledges the good ductility of steel frame structures, the results of post-earthquake reconnaissance and of large-scale shaking table tests have shown that local buckling at the base of the low-storey columns can lead to strength deterioration and exacerbate second-order effects, which are significant contributors to the collapse of steel frame structures. To accurately and efficiently simulate the damage and deterioration behavior of steel frame structures, and to balance the accuracy and efficiency of resilience assessment under strong earthquakes in large-scale computational problems, an MDOF model for steel frame structures incorporating damage and deterioration is proposed. The damage and deterioration mechanisms of steel frame structures are analyzed upon the Japan’s E-Defense shaking table test results. A storey hysteretic model incorporating damage and deterioration is developed using the skeleton shift model. The accuracy and efficiency of the MDOF model proposed are compared with two commonly used planar frame models using data from the collapse test of 18-storey steel frame on the E-Defense shaking table. The research results indicate that, under the earthquake levels considered for seismic resilience assessment of buildings, the MDOF model proposed for steel frame structures incorporating damage and degradation can accurately and efficiently obtain key engineering demand parameters such as inter-storey drift ratios and floor accelerations, with a prediction error of less than 15% and a computational efficiency improvement of over 10 times, providing a highly efficient and accurate computational model for large-scale computational problems.