The performance of fiber-reinforced composite materials is significantly influenced by the mechanical properties of the yarns. Predictive simulations of the mechanical response of yarns are, thus, necessary for fiber-reinforced composite materials. This paper developed a novel experiment equipment and approach to characterize the bending behavior of yarns, which was also analyzed by characterization parameters, bending load, bending stiffness, and realistic contact area. Inspired by the digital element approach, an improved modeling methodology with the probability distribution was employed to establish the geometry model of yarns and simulated bending behavior of yarns by defining the crimp strain of fibers in the yarn and the effective elastic modulus of yarns as random variables. The accuracy of the developed model was confirmed by the experimental approach. More bending behavior of yarns, including the twisted and plied yarns, was predicted by numerical simulation. Additionally, models revealed that twist level and number of plies affect yarn bending properties, which need to be adopted as sufficient conditions for the mechanical analysis of fiber-reinforced composite materials. This efficient experiment and modeling method is meaningful to be developed in further virtual weaving research.