The receptivity of a swept-wing boundary layer to localised surface roughness is studied by means of direct numerical simulations (DNS). The flow case considered is meant to model wind tunnel experiments performed at the Arizona State University by Saric & coworkers. The receptivity amplitude of the crossflow disturbances predicted by the DNS is 40% of that measured in the experiments. The DNS results are then used to evaluate the performance of different receptivity models based on either the parabolised stability equations or the finite Reynolds number theory (FRNT). In general it is found that receptivity amplitudes are well predicted for micron sized roughness elements if non-parallel effects are accounted for.