The numerical weather prediction of radiation fog is challenging, as many models typically show large biases for the timing of the onset and dispersal of the fog, as well as for its depth and liquid water content. To understand the role of physical processes, i.e. turbulence, radia-tion, land-surface coupling, and microphysics, we evaluate the HARMONIE and Weather Research and Forecasing (WRF) mesoscale models for two contrasting warm fog episodes at the relatively flat terrain around the Cabauw tower facility in the Netherlands. One case in-volves a radiation fog that arose in calm anticyclonic conditions, and the second is a radiation fog that developed just after a cold front passage. The WRF model represents the radiation fog well, while the HARMONIE model forecasts a stratus lowering fog layer in the first case and hardly any fog in the second case. Permutations of parametrization schemes for boundary-layer mixing, radiation and microphysics, each for two levels of complexity, have been evaluated within the WRF model. It appears that the boundary-layer formulation is critical for forecasting the fog onset, while for fog dispersal the choice of the microphysics scheme is a key element, where a double-moment scheme outperforms any of the single-moment schemes. Finally, the WRF model results appear to be relatively insensitive to horizontal grid spacing, but nesting deteriorates the modelled fog formation. Increasing the domain size leads to a more scattered character of the simulated fog. Model results with one-way or two-way nesting show approximately comparable results.