Abstract: Under the objective statistical conditions of grain filling ratio and grading curve for grain size,we propose an improved two-step modeling procedure to effectively fill a 2D or 3D non-regular solid body with a mass of random grains.It first partitions the complex solid body into fine regular cells,which provides an intermediate structure for the grain filling and makes the locating of grains much easy.Then,an iterative adjustment process is performed to remove the overlaps between adjacently located grains based on their enclosing boxes.We tested our algorithm by simulating the generation of concrete meso-mechanics models,experimental results show that,compared to other filling method,our algorithm achieves better efficiency and feasibility,it is adaptive to random grains of different shape and different size and gets rid of the convex shape limit of the grains.