Abstract: With the increasing demand for the forming quality of three-dimensional (3D) woven composites in the aerospace field, the influence of yarn deformation during the preforming on the final structural performance has become increasingly critical. Conventional macroscopic models are inadequate for accurately describing meso-scale yarn deformation, while meso-scale models face challenges such as the difficulty in decoupling bending and tensile/compressive deformations and the complexity of characterizing nonlinear bending stiffness. To address these issues, this study developed a meso-scale modeling approach for carbon fiber 3D woven fabrics that explicitly incorporates the bending deformation energy of yarns. A solid-continuous shell hybrid element was developed, and an adjacent-element method was proposed to dynamically update yarn curvature, enabling accurate description of nonlinear bending stiffness. A constitutive model coupling transversely isotropic hyperelasticity with linear elasticity was established. Finally, the model was validated by comparing simulations with experiments, including yarn cantilever bending tests, full-scale three-point bending tests, and hemispherical preforming tests. The results demonstrate that the developed high-fidelity meso-scale model can accurately predict yarn deformation during the forming process of 3D woven fabrics, thereby providing theoretical and computational support for the prediction and control of composite forming quality.