Abstract: The hydraulic conductivity of frozen soil is not only one of the main factors affecting the speed of moisture migration, but also an important model parameter in a large number of frost heave models. However, the study of the hydraulic conductivity for frozen soil is often described by empirical formulas. The calculation results lack a theoretical basis, and the predicted values of different empirical formulas often have enormous deviations. Therefore, it is debatable to use the empirical formula to predict the hydraulic conductivity coefficient of frozen soil. In order to reveal the process of moisture migration, according to the theory of water film thermodynamics at the ice-water interface, this paper points out that the water flow can be regarded as the Darcy flow under the equivalent pressure control in frozen soil. On this basis, the expression of pore water freezing temperature and pore radius is obtained. The capillary bundle theory is applied to the frozen soil and combined with the soil frozen characteristic curve to give a theoretical model for predicting the hydraulic conductivity of frozen soil, Meanwhile, the function of pore radius probability density is eliminated to make this proposed model convenience to calculate and use. The calculated values of this model are compared with the experimental data and empirical formulas of the predecessors. The results show that the pore freezing temperature decreases with the decrease of pore radius, and the temperature drop rate also increases. The freezing temperature reduction rate is significantly accelerated especially when the pore radius is less than 10-6 m. Considering the unfrozen pore water and the unfrozen water film as the migration channel of moisture, in general, the calculated values of this model agree well with the experimental results and are better than the empirical formula, which proves the rationality of the model. The predicted values of the hydraulic conductivity of frozen soil obtained by three empirical formulas are even different by four orders of magnitude for Qinghai-Tibet silty clay. Although the predicted value obtained by the Fowler formula is closer to the measured value, there is an empirical coefficient in the prediction formula. For the empirical coefficient is often arbitrarily determined in a certain range, and it is no clear physical meaning and calculation procedure, which leads to the reliability of the predicted value from the empirical formula is debatable. At the same time, the power function is used to fit the soil frozen characteristic curve (SFCC) in this model, which tends to infinity near 0 ℃. In order to ensure the continuity of the function of the SFCC, the piecewise function is used to describe the soil frozen characteristic curve, but the deviation of the calculated and measured values of the model within 0~-0.0357 ℃. Because the model is based on a quasi-steady state process, and the frozen soil is in the stage of intense phase change near 0 ℃. Thus, this paper pointed out that the fitting formula of the soil frozen characteristic curve is very important to this model for reducing this deviation near 0 ℃.