The performance of the UK Meteorological Office operational numerical weather prediction system in the Antarctic is examined. The analysis/forecast model currently has a reasonable representation of the atmospheric circulation, although there are some problems. During the first three days of the forecasts the katabatic winds in the coastal region become too strong compared to those in the initial fields, which results in coastal easterly winds that are too strong. However, the pattern of katabatic winds across the continent is broadly correct, suggesting that there are no major errors in the model orography used. The forecast tropospheric temperatures are generally too cold, although positive errors are found inland of the coast in the region of descent associated with the katabatic wind circulation. The mean forecast low pressure centres within the circumpolar trough are in approximately the correct locations, although the depth of the centres is too deep. During January there is good agreement between the model total cloud amount over the Antarctic and the estimates of total cloud cover from in situ observations. However, in July the model has too much cloud. There is no significant change in the amount of cloud during the first three days of the forecasts. The model has a good representation of the broadscale field of snow accumulation across the Antarctic. A comparison of various 6 hour forecast model fields against observations confirms the errors noted in the three-day forecasts when compared against analyses