Abstract Six cases of two-week numerical weather prediction experiments, begun from and verified against actual data, are presented to illustrate the extended-range forecasting capability of a global circulation model. The forecasts, all for Northern Hemisphere winter, are analyzed for both transient and time-mean properties of the predicted fields of wind, temperature, pressure, and precipitation. Rms temperature and sea-level pressure errors rise above persistence level during the first week, but forecast tropospheric zonal winds and 500 mb heights are superior to persistence throughout the two-week period. Time-mean forecasts display the model's climatological bias, but show skill in the prediction of surface temperature and synoptic-scale circulation patterns representing an improvement over climatology. Skill in precipitation forecasting is demonstrable for about one week.