Soil water migration is a significant factor in the development of subgrade ice layers in permafrost areas. The prediction of moisture inflow to the freezing zone is an important element in the design and analysis of robust highway subgrade in permafrost regions. In order to better understand moisture inflow to the freezing zone, we designed an experimental investigation to monitor the variation of water content and temperature in freezing soil. Identical experiments were conducted using three different soil types: clay, silt, and fine sand. Moisture was supplied from the sample base while the column was maintained at a constant nonfreezing temperature and moisture equilibration was achieved. A temperature gradient was then applied to the sample via the application of a subfreezing temperature at the column surface. The changes in the temperature and water content of the sample were measured at regular time intervals. Based on the freezing rate, the freezing process can be classified into three stages: the quick frost stage, the transition frost stage, and the stable frost stage. During the freezing process, the inflow rates increased as the thickness of the ice lens increased. When the maximum rate was reached, the final (maximum) thickness of the ice lens was attained. Subsequently, the water inflow rates decreased. All of the water supplied from the bottom of the sample flowed into the frost section during the freezing process, with the moisture contents in the lower portion remaining relatively unchanged. The segregation potential changed with the freezing rate and soil type. This paper proposes the concept of “generalized segregation potential” to extend the traditional segregation potential concept. The use of this new concept with an existing moisture inflow prediction model provided excellent correspondence to measured inflow rates for all three study soils in the early and late stages of the test but overpredicted the inflow rates in the mid-range of the test.