The predictive capability of a three-dimensional (3D) numerical model for sediment transport and resulting scour around a structure is investigated in this study. Starting with the bed-load and suspended-load sediment transport (reference) model developed by Takahashi (2000, “Modeling Sediment Transport Due to Tsunamis With Exchange Rate Between Bed Load Layer and Suspended Load Layer,” Proceedings of the 27th International Conference on Coastal Engineering, ASCE, Sydney, Australia, pp. 1508–1519), we first introduce an extension to incorporate Nielsen’s modified Shields parameter to account for the effects of infiltration/exfiltration flow velocity across the fluid-sand interface on the sediment transport (the modified Shields-parameter model). We then propose a new model to include the influence of the effective stress to account for the stress fluctuations inside the surface layer of the sand bed (the effective-stress model). The three analytical models are incorporated into a 3D numerical solver developed by Nakamura (2008, “Tsunami Scour Around a Square Structure,” Coast. Eng. Japan, 50(2), pp. 209–246) to analyze the dynamics of fluid-sediment interaction and scour. Their solver is composed of two modules, namely, a finite-difference numerical wave tank and a finite-element coupled sand-skeleton pore-water module. The predictive capability of the three alternative coupled models is calibrated against hydraulic experiments on sediment transport and resulting scour around a fixed rigid structure due to the run-up of a single large wave in terms of the sediment transport process and the final scour profile after the wave run-up. It is found that, among the three models considered, the proposed effective-stress model most accurately predicts the scouring process around the seaward corner of the structure. The results also reveal that the deposition and erosion patterns predicted using the effective-stress model are in good agreement with measured results, while a scour hole at the seaward corner of the structure cannot be always predicted by the other two models.