Groundwater variation is a significant cause for high-fill foundation settlement. In this study, centrifugal model tests were carried out to study the settlement of a high-fill foundation using a wetting–drying system for simulating water level variations. Three scaling methods including equivalent substitution method, trapezoid similarity method and parallel gradation method, were utilised to prepare the particle-size distribution of tested specimens. The test process consists of four loading cycles and each loading cycle consists of construction, water rise and drainage stages. The analysis results showed that settlement after construction only accounted for less than 20% of the total settlement. Besides, the settlement of the backfill increases with the filling height. The ratio between the settlement and the filling height of the specimen prepared by the equivalent substitution method is higher than the field monitoring data, while those prepared by the trapezoid similarity method or parallel gradation method agree well with the measurements. The settlement after construction with time can be predicted by the three-point method. Furthermore, significant settlement can be observed during the first wetting-drying cycle. The settlement during the water level rise was remarkably greater than that during the draw-down of the water level in each wetting–drying cycle.