Abstract: Reconstruction of the porous media model is the basis and key to the study of the meso-seepage mechanism of soil. The soil models are constructed by the quartet structure generation set (QSGS), and the lattice Boltzmann method (LBM) is used to examine the meso-seepage mechanism of the reconstructed soil under different conditions through MATLAB self-programming. The results show that the degree of pore connectivity increases significantly as the model size increases. The 300×300 grid size model has the largest increase in the connected porosity (34.38%). If the model size continues to be expanded, the increase is not obvious. Fluid particles can form the main seepage channel in the area with good pore connectivity and large pore size, and there is a fingering effect. The flow velocity in the middle of the pore channel is the largest, up to 0.0324. The flow velocity is getting smaller and smaller when it gets closer and closer to the pore wall. The velocity of the large porosity soil is larger than that of the small porosity soil, and the velocity of the low porosity soil is more stable than that of the large porosity soil. Comparing the calculation results of LBM simulation permeability with those of the typical K-C formula, it is found that the higher the porosity, the more accurate the calculation of permeability (n=0.78, the error is 10.22%). The smaller the soil particles, the narrower the seepage channels and the denser the distribution. The corresponding velocity field distribution is more uniform, and the flow velocity is also smaller. The research results can better reveal the meso-seepage mechanism of the reconstructed soil, and can also provide a certain reference for the existing meso-soil pore flow research methods.