An accurate prediction of wind-induced redistribution of snow load on roof surfaces is vital to structural design. To represent the pattern of snow distribution caused by snowdrift in wind tunnel test, appropriate modeling particles should be selected. The particle density is the key to determine the values of several important similarity parameters. In this study, the redistribution of snow load on a stepped flat roof was simulated by means of wind tunnel test using low-density saw wood ash, medium-density polyfoam, and high-density silica sand, respectively. To ensure the comparability of the test results of the three modeling particles, the wind tunnel test results for comparison were performed under almost the same conditions of dimensionless wind velocity and dimensionless time. Then, the results of the present study were compared with those from field observations of prototypes in previous studies. The effects of wind duration, wind velocity, and roof span on the redistribution of snow on roof surfaces were investigated. The characteristics of erosion/deposition range and the location of maximum quantities of erosion/deposition under independent effects of wind duration, wind velocity, and roof span were also studied.