To analyse the applicability of a limited-area atmosphere model to the Southern Ocean, a one-year simulation for 1985 is performed using the REgional MOdel REMO at 55-km horizontal grid-spacing implemented for the Antarctic regions of the Weddell, Bellingshausen and Amundsen Seas. To evaluate the performance of REMO, a comparison of model results to observations and to reanalysis/analysis data sets is carried out. REMO is initialized and driven at the lateral and lower boundaries by data of the European Centre for Medium-Range Weather Forecasts (ECMWF) reanalysis (ERA 15). Overall, REMO is an appropriate tool for further climate studies in Antarctic regions. It reproduces reasonably well basic spatial patterns and the seasonal cycle of the atmospheric circulation. However, the simulated mean sea level pressure (MSLP) is predominantly lower than the MSLP provided by observations and by ERA. Considerable temperature differences in the lower troposphere over sea ice in winter cause discrepancies between the REMO and ERA pressure fields in the mid-troposphere too. The precipitation rate P of the REMO simulation agrees qualitatively well with main features of the observed climatological spatial distribution described in literature. The seasonal cycle of P in the inner Weddell Sea reflects the Antarctic semi-annual oscillation. Concerning the forcing fields, the ERA sea ice surface temperatures in winter are generally higher than satellite derived surface temperatures. Although the differences are 10 to 15 K in the southern Weddell Sea, this deficiency of the ERA data hardly influences the mean large-scale circulation.