To investigate the variations in friction in underground silos due to groundwater during backfilling, three silo models, namely, A, B, and C, with three different diameters of 400 mm, 500 mm, and 600 mm were designed. All these models were composed of 2 mm thick steel plates with a bottom inclination angle of 35°. Six water injection tests were conducted under two different backfilling conditions. The experimental results demonstrate that when there are sudden displacement changes at the top of the silos, the water levels of models A, B, and C are 41.40 mm, 39.80 mm, and 35.00 mm, respectively, under backfilling condition a. Meanwhile, the water levels become 40.60 mm, 39.00 mm, and 33.10 mm under backfilling condition b (p<0.05). In the stationary phase, the reaction forces at the top of the models A, B, and C remain unchanged. In the floating phase, the reaction force at the top of the silos abruptly changes. Under backfilling condition a, the water level-reaction force curves of models A, B, and C result in inflection points at water levels of 37.10 mm, 27.00 mm, and 21.00 mm, respectively. Under condition b, the inflection points are observed at water levels of 37.10 mm, 30.05 mm, and 24.00 mm, respectively (p<0.05). Because of the static frictional force, the displacement changes measured by the gauge lag behind the variation in reaction forces. In the stationary phase, under condition a, the maximum static frictional force of the models is 206.17 N, 316.81 N, and 364.16 N, respectively. These values are all greater than those under condition b, which are 197.61 N, 310.82 N, and 352.96 N, respectively; yet, the water levels of the models under two different conditions increase at nearly the same rate (p<0.05). In the floating phase, frictional force reduces rapidly under both conditions and then levels off gradually. In actual engineering projects, favorable anti-floating underground silo designs can be adopted by integrating the frictional force variations in the floating phase.