Grain shear stress has been used in many earlier studies to estimate the sediment transport capacity of interrill overland flow. However, this procedure may be flawed because turbulent eddies generated by large-scale roughness elements contribute to sediment transport in this type of flow, whereas they do not in deeper (river) flow. In this study a procedure is developed for estimating the proportion of total shear stress contributing to sediment transport in overland flow. Application of this procedure to 1506 flume experiments representing a wide range of discharge, slope and surface roughness conditions reveals that the proportion of total shear stress contributing to sediment transport decreases from 1.0 to 0.83 as roughness concentration increases from 0 to 0.37. Concomitantly, the grain shear stress expressed as a proportion of total shear stress decreases from 1.0 to 0.14. Clearly, the proportion of total shear stress involved in sediment transport is higher than is indicated by grain shear stress. Consequently, use of grain shear stress to predict the sediment transport capacity of overland flow on rough surfaces will result in significant underestimation of the transport capacity.