For the efficient shape analysis of massive, heterogeneous and complex 3D models, an optimization method for 3D model shape based on optimal minimum spanning tree was proposed. Firstly, a model description based on 3D model Minimum Spanning Tree (3D-MST) was constructed. Secondly, local optimization was realized by topology and geometry detection and combination of bilateral filtering and entropy weight distribution, obtaining optimized MST representation of the model. Finally, the shape analysis and similarity detection of the model were realized by optimized Laplacian spectral characteristics and Thin Plate Spline (TPS). The experimental results show that the proposed method not only effectively preserves shape features of the model, but also effectively realizes sparse optimization representation of the complex model, improving the efficiency and robustness of geometric processing and shape retrieval.