Soil salinization is a crucial issue in arid and semi-arid regions, especially for reservoir-based irrigation systems. In this study, the HYDRUS-1D model was used to investigate soil salinity resulting from seepage from a lowland reservoir in Xinjiang province, China. After successfully developing a model to simulate field observations, it was used to evaluate four hypothetical scenarios with different groundwater (GW) salinities and depths to GW. The model was calibrated, validated, and run for scenarios with periods of 367, 363, and 436 days. Root mean square error values of soil water and soil salinity ranged from 0.000 to 0.053 cm3/cm3 and 0.56 to 5.66 g/L, respectively, during calibration, and from 0.009 to 0.039 cm3/cm3 and 1.86 to 7.52 g/L, respectively, during validation. The results indicate that soil salinity downstream of the reservoir depends strongly on the depth to the GW level, while GW salinity has a much smaller impact. Controlling leakage from lowland reservoirs to avoid rising GW levels is therefore important to minimize soil salinization. These findings are generally useful for lowland reservoir design and construction, and for irrigation management in arid regions.