Abstract: The research on homogenization of ordinary masonry has been relatively developed, but the research on historical masonry with random geometry and material is lacking. Based on the method of finite-scale test windows, a method for selecting a representative volume element (RVE) of masonry structures is proposed and is proved to be correct by comparisons with the results of published experiments and conventional finite element models (FEM). An appropriate RVE for Tibetan ancient stone masonry structures is selected. The influence of the size of RVEs and the distribution of components on the effective modulus are investigated. Based on the selected RVE, a homogenization model and a macro model of the Tibetan stone masonry structure are established. The results indicate that the selecting method, which is suitable for periodic and quasi-periodic masonry structures, can select an RVE with the mechanical properties close to the complete structure. As the size of the RVE becomes larger, its Voigt and Reuss effective modulus gradually converge to the modulus of the complete structure and present a trend of rapid and slow progresses. The difference in the component distribution will change the convergence amplitude of the effective modulus, but on a larger RVE, the influence of the component distribution will be offset by that of the size of the RVE. The homogeneous model can replace the conventional FEM to analyze local structures and give the rule of the stress distribution of Tibetan ancient stone masonry structures. The macro model can replace the conventional FEM to simulate the macro deformation of an overall structure.