Soil surface crusts are widely reported to favour Hortonian runoff, but are not explicitly represented in most rainfall-runoff models. The aim of this paper is to assess the impact of soil surface crusts on infiltration and runoff modelling at two spatial scales, i.e. the local scale and the plot scale. At the local scale, two separate single ring infiltration experiments are undertaken. The first is performed on the undisturbed soil, whereas the second is done after removal of the soil surface crust. The HYDRUS 2D two-dimensional vertical infiltration model is then used in an inverse modelling approach, first to estimate the soil hydraulic properties of the crust and the subsoil, and then the effective hydraulic properties of the soil represented as a single uniform layer. The results show that the crust hydraulic conductivity is 10 times lower than that of the subsoil, thus illustrating the limiting role the crust has on infiltration. Moving up to the plot scale, a rainfall-runoff model coupling the Richards equation to a transfer function is used to simulate Hortonian overland flow hydrographs. The previously calculated hydraulic properties are used, and a comparison is undertaken between a single-layer and a double-layer representation of the crusted soil. The results of the rainfall-runoff model show that the soil hydraulic properties calculated at the local scale give acceptable results when used to model runoff at the plot scale directly, without any numerical calibration. Also, at the plot scale, no clear improvement of the results can be seen when using a double-layer representation of the soil in comparison with a single homogeneous layer. This is due to the hydrological characteristics of Hortonian runoff, which is triggered by a rainfall intensity exceeding the saturated hydraulic conductivity of the soil surface. Consequently, the rainfall-runoff model is more sensitive to rainfall than to the subsoil's hydrodynamic properties. Therefore, the use of a double-layer soil model to represent runoff on a crusted soil does not seem necessary, as the increase of precision in the soil discretization is not justified by a better performance of the model.