Numerical solutions could not perform rapid system-level simulation of the behavior of micro-electro-mechanical systems(MEMS) and analytic solutions for the describing partial differential equations are only available for simple geometries.Model order reduction(MOR) can extract approximate low-order model from the original large scale system.Conventional model order reduction algorithm is based on first-order system model,however,most structure mechanical MEMS systems are naturally second-order in time.For the purpose of solving the above problem,a direct second-order system model order reduction approach based on Krylov subspace projection for the coupled dynamic study of electrostatic torsional micromirrors is presented.The block Arnoldi process is applied to create the orthonormal vectors to construct the projection matrix,which enables the extraction of the low order model from the discretized system assembled through finite element analysis.The transfer functions of the reduced order model and the original model are expanded to demonstrate the moment-matching property of the second-order model reduction algorithm.The torsion and bending effect are included in the finite element model,and the squeeze film damping effect is considered as well.An empirical method considering relative error convergence is adopted to obtain the optimal choice of the order for the reduced model.A comparison research between the full model and the reduced model is carried out.The modeling accuracy and computation efficiency of the presented second-order model reduction method are confirmed by the comparison research results.The research provides references for MOR of MEMS. Numerical solutions could not perform rapid system-level simulation of the behavior of micro-electro-mechanical systems (MEMS) and analytic solutions for describing partial differential equations are only available for simple geometries. Model order reduction (MOR) can extract approximate low- order model from the original large scale system. Conventional model order reduction algorithm is based on first-order system model, however, most structure mechanical MEMS systems are naturally second-order in time. For the purpose of solving the above problem, a direct second -order system model order reduction approach based on Krylov subspace projection for the coupled dynamic study of electrostatic torsional micromirrors is presented. The block Arnoldi process is applied to create the orthonormal vectors to construct the projection matrix, which enables the extraction of the low order model from the discretized system assembled through finite element analysis. transfer functions of the reduced or der model and the original models are expanded to demonstrate the moment-matching property of the second-order model reduction algorithm. torsion and bending effect are included in the finite element model, and the squeeze film damping effect is considered as well. An empirical method considering relative error convergence is obtained to obtain the optimal choice of the order for the reduced model. A comparison research between the full model and the reduced model is carried out. The modeling accuracy and computation efficiency of the presented second-order model reduction method are confirmed by the comparison research results. The research provides references for MOR of MEMS.