The invention discloses a method for building graded broken stone shear resistance number value prediction models. The method includes the steps: (1) selecting sample particles; (2) virtual modeling of coarse aggregate particles: acquiring contour coordinates of the sample particles and building virtual particle matrix samples to form large-batch particles; (3) virtual modeling of shear test parts: building wall models to simulate test models, and removing error escape particles in a screened manner; (4) virtual shear model loading: giving a servo operation mechanism and a constant-speed operation mechanism to the wall models in different stages to simulate stress and strain control modes in a shear-resistant test; (5) test result counting and arrangement: recording stress and strain curves in shear-resistant test processes and stress and strain data in the limit breaking state. According to the method, the graded broken stone shear resistance number value prediction models are built in a batch manner, numerical simulation of shear-resistant evaluation is achieved, and shear-resistant characteristics of graded broken stones can be rapidly evaluated.