The modeling of the atmospheric boundary layer over sea ice is still challenging because of the complex interaction between clouds, radiation and turbulence over the often inhomogeneous sea ice cover. There is still much uncertainty concerning sea ice roughness, near-surface thermal stability and related processes, and their accurate parameterization. Here, a regional Arctic climate model forced by ERA-interim data was used to test the sensitivity of climate simulations to a modified surface flux parameterization for wintertime conditions over the Arctic. The reference parameterization as well as the modified one is based on Monin–Obukhov similarity theory, but different roughness lengths were prescribed and the stability dependence of the transfer coefficients for momentum, heat and moisture differed from each other. The modified parameterization accounts for the most compre- hensive observations that are presently available over sea ice in the inner Arctic. Independent of the parameterization used, the model was able to reproduce the two observed dominant winter states with respect to cloud cover and longwave radiation. A stepwise use of the different parameterization assumptions showed that modifications of both surface roughness and stability dependence had a considerable impact on quantities such as air pressure, wind and near-surface turbulent fluxes. However, the reduction of surface roughness to values agreeing with those observed during t he Surface Heat Budget of the Arctic Ocean campaign led to an improvement in the western Arctic, while the modified stability parameteri- zation had only a minor impact. The latter could be traced back to the model's underestimation of the strength of stability over sea ice. Future work should concentrate on possible reasons for this underestimation and on the question of generality of the results for other climate models