This paper presents an innovative study on the dimension variation prediction and control for polymer matrix fiber reinforced composites. A dimension variation model was developed for process simulation based on thermal stress analysis and finite element analysis (FEA). This model was validated against the experimental data, the analytical solutions and the data from literature. Using the FEA-based dimension variation model, the deformations of typical composite structures were studied and the regression-based dimension variation model was developed. By introducing the material modification coefficient, this comprehensive model can account for various fiber/resin types and stacking sequences. The regression-based dimension variation model can significantly reduce computation time by eliminating the complicated, time-consuming finite element meshing and material parameter defining process, which provides a quick design guide for composite products with reduced dimension variations. The structural tree method (STM) was developed to compute the assembly deformation from the deformations of individual components, as well as the deformation of general shape composite components.The STM enables rapid dimension variation analysis/synthesis for complex composite assemblies with the regression-based dimension variation model. The exploring work presented in this research provides a foundation to develop practical and proactive dimension control techniques for composite products.