A cross-scale analysis method combining macro and meso scales was adopted to establish a unit cell model of plain-woven SiCf/SiC ceramic matrix composites. Random pores were introduced into the matrix to predict the effective thermal expansion coefficient. The study developed a macro-meso multiscale model of yarn units and woven structures, and analyzed the thermal expansion responses of yarns and woven unit cells under different temperatures through finite element simulations. Experimental comparisons demonstrated the feasibility of using multiscale methods to predict the material’s thermal expansion coefficient. By calculating thermal expansion coefficients of unit cell models with different porosity levels, a relationship curve between porosity and thermal expansion coefficient in randomly distributed pore models was obtained. Results indicate that for plain-woven SiCf/SiC ceramic matrix composites, the randomly distributed pore structure in the matrix has minimal influence on the overall thermal expansion coefficient of the material.