Abstract: To study RCC compaction mechanisms from the mesoscopic perspective, inversion of meso-structural parameters was calculated by modelling uniaxial compression tests of fresh roller-compacted concrete (RCC) using the Particle Flow Code (PFC). First, basing on sieving statistical results, we developed a method for determination of the number of particles with different sizes so as to simulate the mixture proportion and aggregate gradations of RCC, and simulated the geometric shapes of irregular aggregates using a 3D laser scanning technology. Then, a new method for PFC simulation of RCC uniaxial compression tests was developed, and a BP neural network was established to replace complicated PFC simulations. Finally, we present an inversion method of RCC meso-structural parameters along with its solution algorithm of adaptive differential evolution (ADE), which takes the stress-strain curve of a compression test as its calibration target. A case study shows that the uniaxial compression curves, achieved from the PFC simulation using the inversed parameters, agree well with the test curves. The PFC model and inversed meso-parameters presented in this paper would lay a theoretical basis for further studies on RCC vibrating compaction and its mesoscopic mechanism of layers bonding.