Abstract: As a novel type of concrete material, the mesoscopic structural characteristics of rock-filled concrete (RFC) are significantly different from those of conventional concrete. Therefore, the accurate construction of its mesoscopic mechanical model is of great theoretical significance and of practical engineering value. This study presents a novel 3D mesoscopic numerical model for rock-filled concrete, which accurately represents the mesoscopic physical characteristics of the material. First, an improved random aggregate model generation strategy based on Laguerre tessellation is proposed, which iteratively generates polyhedral random aggregates with adjustable volume fractions and gradation. Next, an automated strategy for constructing the self-supporting rock skeleton is developed, which optimizes re-meshing produce to create realistic contact surfaces among aggregates. The final model includes three phases: self-compacting concrete, rock, and interface transition zones, with adjustable rockfill ratios. Using this model, 13 RFC and 60 conventional concrete numerical models are generated to investigate the effects of aggregate volume fraction and porosity on the elastic properties of concrete. The research results show that the reduction in elastic modulus of RFC with increased porosity is relatively small. Additionally, the model is used for simulating the uniaxial compression of RFC, and the numerical stress-strain curve matches the experimental results. The mesoscopic numerical model proposed effectively captures the irregular, angular characteristics of rocks and explicitly represents the contact surfaces among the rock skeletons. It can be used for predicting the entire process from the elastic stage to nonlinear damage in RFC.