Infiltration into frozen soil is a result of the whole climate dynamics of the preceding winter with all its importance for the freezing of the soil. Therefore a predictive infiltration model needs to include a proper description of the main processes of soil water and heat transfer during season-long periods. Such a model may assume two water-conducting flow domains. A lysimeter experiment was set up with the aim of studying these processes in two different sandy soils. Frequent measurements of total and liquid soil water content, soil temperature, and groundwater level were made during two winters with contrasting meteorological conditions. The main problems in the simulation of the two winters were (1) frost-induced upward water redistribution, (2) rate of infiltration in the initially air-filled pores, and (3) heat transfer caused by snowmelt refreezing in the frozen soil. An extensive calibration of the model suggested that some key empirical parameters were not constant for the two soils and the two seasons. Complementary methods for determining the hydraulic conductivity of frozen unsaturated field soils are necessary to further improve the model. ¿ 1999 American Geophysical Union