Motivation Solar wind ion precipitation on airless rocky bodies like the Moon has been linked to a variety of processes summarised by the term space weathering, including darkening and a red shift in optical reflectance spectra, the formation of vesicles and implantation of ions in mineral grains and the formation of an amorphised rim therein [1]. Solar wind irradiation additionally leads to the release of surface material via the kinetic sputtering process. These ejecta undergo ballistic trajectories and take part in the formation of the Lunar exosphere, which is well documented by space missions and ground-based observations. To properly gauge the importance of the sputtering contribution, however, a more detailed physical implementation of the sputtering process is necessary than used so far in, e.g., [2]. Prior, this has typically been approximated by simulation codes using the binary collision approximation (BCA) like SRIM [3], see also the review in [4]. On the other hand, more recent literature suggests significant limitations of this software [5,6] – especially related to real-life surfaces like lunar regolith. We present a combined experimental and numerical work on the sputtering yields of lunar regolith and discuss the validity of popular BCA codes. Methods Experimental data on sputtering yields, i.e., the amount of material released normalised per number of incidence ions, were obtained using a quarz crystal microblance (QCM) setup [7]. To overcome limitations on the sample configuration that come with this approach, an additional QCM served as a catcher for the sputtered material [8]. This allowed us to conduct experiments on rough bulk samples and additionally probe the ejecta angular distribution. We perform such experiments with lunar regolith (from Apollo 16 sample #68501) and helium and hydrogen ions at solar wind energies of 1 keV/nucleon. Numerical investigations were carried out both using SRIM and variants of SDTrimSP in its 1D and 3D version [9]. The latter in particular enables systematic investigations with varying surface roughness comparable to the experiments, as well as surface porosity resembling the regolith structures observed on the Moon [10]. Results In line with previous studies, a comparison for flat samples reveals that the BCA approach consistently overestimates sputtering yields, with SRIM deviating the most [6]. By introducing surface roughness through consideration of regolith grains, the sputtering yield as a function of incidence angle is both flattened and reduced. Both effects become even more pronounced when surface porosity is included in the description. Three-dimensional simulations match these results qualitatively and, when the initial offset is factored out, also quantitatively. We will present these findings in greater detail and discuss possible implications in the context of the lunar exosphere.