In this article, two new model reduction methods for fast dynamic performance optimization of complicated beam-type structures are presented. In the new model reduction methods the projection of structural nodal degrees of freedom is carried out using the localized base vectors, which are obtained by physical hypothesis. The first new model reduction method constructs the base vectors based on the assumption of block-wise rigid body motion modified by minimizing the displacement error measured in strain energy. The second new model reduction method is based on the plane cross-section assumption, and the crosssectional warp deformation is also taken into consideration. Then, two reduced optimization models based on each method and design variable decomposition are formulated and solved. Both reduced optimization models are compared with the response surface method-based structural optimization approach. Some examples are presented to demonstrate the feasibility and effectiveness of the proposed methods.