Intensive agricultural practices on sensitive soils induce high erosion rates in central Belgium. Expert-rules models quantify runoff and erosion at catchment scale, avoiding over-parameterization, and can include some direct or indirect connectivity features. The aim of this article is to test the ability of an expert-based model, LandSoil, to quantify runoff and to locate erosion and sedimentation areas in a small cultivated loamy catchment in Belgium during the years 2014, 2015 and 2016. Spatialized data are important for assessing model outputs and the erosive response. Measurements of runoff and observation of spatial erosion/deposition patterns, especially around major connectivity points, permitted an assessment of the reliability of the model results. Runoff modelling gave contrasting results (good linear adjustment at the outlet of the 83 ha sub-catchment (point 1): r2 of 0.96, Nash–Sutcliffe criterion of 0.95; less good at the outlet of the 3.9 ha sub-catchment (point 2): r2 of 0.28, Nash–Sutcliffe criterion of –0.47). For point 2 the poor results are explained by the very few runoff events observed, a scaling effect and the small area with a single land use. Graduated rulers demonstrate that the model is able to provide a coherent pattern of erosion/deposition. The study highlights great sensitivity to the effect of land use, land allocation, landscape design and slope gradients. Grass strips induce deposition of eroded particles when slopes are gentle (< 2%). Woodland strips decrease connectivity by being in the stream but deposit thinner sediment layers. Field boundaries have a role in the transport, but not really the quantity, of sediments. This model validation in the Belgian loess context allows us to use LandSoil in other similar environments in order to estimate the effects of landscape management scenarios.