Abstract: The hydraulic conductivity of frozen soils macroscopically reflects the migration and phase change of liquid in soil, which is the key to the phenomenon of frost-heave. The high sensitivity of frozen soil to temperature and fluid presents significant challenges for the physical measurement of its hydraulic conductivity. Therefore, the development of reliable numerical analysis methods at micro level is a valuable exploration. In this study, the structure of the porous medium of soil samples is first reconstructed using gradation curves, and then a numerical method based on PFM (Phase Field Method) and LBM (Lattice Boltzmann Method) coupling is proposed to investigate the permeability characteristics of frozen soil at the mesoscale. Based on existing experimental data and analytical models, the proposed model is validated. On this basis, the effects of ice content and different ice formation methods on the permeability characteristics of frozen soils are studied. The results show that the ice content has a significant effect on the hydraulic conductivity of frozen soils, which is mainly reflected in the increase of the fluid relative specific interface length Rf and the tortuosity τf. The effect of pore filling and particle coating on the permeability coefficient of frozen soil is different: the former is more closely related to Rf; for the latter, the influence of τf becomes stronger with increasing ice content.