Abstract. Katabatic winds are a key feature of the climate of Antarctica, but despite decades of extensive studies, substantial biases remain in their representation in atmospheric models. However, it is often difficult to identify the origin of those biases amongst the model resolution, physical content, and large-scale forcings aspects. This study conducts an extensive investigation of the ability of the ICOLMDZ atmospheric model to simulate Antarctic katabatic winds by disentangling uncertainties associated with parameter calibration, from those associated with horizontal resolution as well as structural deficiencies in the model with a particular attention given to turbulent diffusion. We carefully select a katabatic-driven wind event in clear-sky conditions in Adélie Land, and perform perturbed parameter experiments at three different horizontal resolutions (10, 20 and 40 km). ICOLMDZ is able to reproduce wind observations, but the parametric uncertainty remains large and structural differences not associated with parameter calibration nor horizontal resolution are found for turbulence and near-surface temperature. A parametric analysis reveals that the most critical parameter controlling the magnitude of near-surface winds is roughness length, whereas near-surface temperatures are mainly controlled by snow near-infrared albedo. Sensitivity to horizontal resolution reveals that the 40 km configuration compares least favourably with the observations, and that the 10 and 20 km ensembles cannot be distinguished due to a too wide parametric uncertainty. We then discuss three aspects of katabatic winds modeling that we deem critical but underappreciated: the parameterization of roughness length over snow, oscillations in katabatic flows, and the representation of subgrid-scale orographic drag. This study underlines in particular the need for a more physical parameterization of roughness length to correctly represent near-surface wind along the slopes of Antarctica.