Two-week predictions were made for two winter cases by applying the Geophysical Fluid Dynamics Laboratory high-resolution, nine-level, hemispheric, moist general circulation model. Three versions of the model are discussed: Experiment 1 includes the orography but not the radiative transfer or the turbulent exchange of heat and moisture with the lower boundary; Experiment 2 accounts for all of these effects as well as land-sea contrast; Experiment 3 allows, in addition, the difference in thermal properties between the land-ice and sea-ice surfaces, as well as an 80% relative humidity condensation criterion reduced from the 100% criterion in Experiments 1 and 2. The computed results are compared with observed data in terms of the evolution of individual cyclonic and anticyclonic patterns, the zonal mean structure of temperature, wind, and humidity, the precipitation over the United States, and the hemispheric energetics. The forecast near sea level was considerably improved in Experiments 2 and 3 over Experiment 1. The experiment succeeded in forecasting the birth of second and third generation extratropical cyclones and their behavior thereafter. The hemispheric sum of precipitation was increased five times in Experiment 2 over that in Experiment 1, and even more in Experiment 3, the greatest contribution occurring in the Tropics. Two winter cases were considered. The correlation coefficients between the observed and the forecast patterns for the change of 500-mb geopotential height from the initial time remained above 0.5 for 13 days in one case and for 9 days in the other. There are, however, several defects in the model. The forecast temperature was too low. In the flow pattern the intensities of the Highs and Lows weakened appreciably after 6 or 8 days, reflecting the fact that the forecast of eddy kinetic energy was less than the observed. On the other hand, the intensity of the tropospheric westerlies was too great.