The long-term mean climatology obtained from integrations conducted with different resolutions of the GFDL “SKYHI” finite-difference general circulation model is examined. A number of improvements that have been made recently in the model are also described. The versions considered have 3° × 3.6°, 2° × 2.4°, and 1° × 1.2° latitude–longitude resolution, and in each case the model is run with 40 levels from the ground to 0.0096 mb. The integrations all employ a fixed climatological cycle of sea surface temperature. Over 25 years of integration with the 3° model and shorter integrations with the higher-resolution versions are analyzed. Attention is focused on the December–February and June–August periods. The model does a reasonable job of representing the atmospheric flow in the troposphere and lower stratosphere. The simulated tropospheric climatology has an interesting sensitivity to horizontal resolution. In common with several spectral GCMs that have been examined earlier, the surface zonal-mean westerlies in the SKYHI extratropics become stronger with increasing horizontal resolution. However, this “zonalization” of the flow with resolution is not as prominent in the upper troposphere of SKYHI as it is in some spectral models. It is noteworthy that—without parameterized gravity wave drag—the SKYHI model at all three resolutions can simulate a realistic separation of the subtropical and polar night jet streams and a fairly realistic strength of the lower-stratospheric winter polar vortex. The geographical distribution of the annual-mean and seasonal precipitation are reasonably well simulated. When compared against observations in an objective manner, the SKYHI global precipitation simulation is found to be as good or better than that obtained by other state-of-the-art general circulation models. However, some significant shortcomings remain, most notably in the summer extratropical land areas and in the tropical summer monsoon regions. The time-mean precipitation simulation is remarkably insensitive to the horizontal model resolution employed. The other tropospheric feature examined in detail is the tropopause temperature. The whole troposphere suffers from a cold bias of the order of a few degrees Celcius, but in the 3° SKYHI model this grows to about 6°C at 100 mb. Interestingly, the upper-tropospheric bias is reduced with increasing horizontal resolution, despite that the cloud parameters in the radiation code are specified identically in each version. The simulated polar vortex in the Northern Hemisphere winter in the upper stratosphere is unrealistically confined to high latitudes, although the maximum zonal-mean zonal wind is close to observed values. Near the stratopause the June–August mean temperatures at the South Pole are colder than observations by ∼65°C, 50°C, and 30°C in the 3°, 2°, and 1° simulations, respectively. The corresponding zonal-mean zonal wind patterns display an unrealistically strong polar vortex. The extratropical stratosphere stationary wave field in the Northern Hemisphere winter is examined in some detail using the multiyear averages available from the 3° SKYHI integration. Comparison with comparable long-term mean observations suggests that the model captures the amplitude and phase of the stationary waves rather well. The SKYHI model simulates the reversed equator-pole temperature gradient near the summer mesopause. The simulated summer polar mesopause temperatures decrease with increasing, horizontal resolution, although even at 1° resolution the predicted temperatures are still warmer than observed. The increasing resolution is accompanied by increased westerly driving of the mean flow in the summer mesosphere by dissipating gravity waves. The present results suggest that the SKYHI model does explicitly resolve a significant component of the gravity waves required to produce the observed summer mesopause structure. The semiannual oscillation near the tropical stratopause is reasonably well simulated in the 3° version. The main deficiency is in the westerly phase, which is not as strong as observed. There is also a second peak in the amplitude of the semiannual wind oscillation at the top model level (0.0096 mb) corresponding to the observed mesopause semiannual oscillation. This simulated mesopause oscillation is weaker (by a factor of ∼3) than that observed. The simulation in the tropical stratopause and mesosphere changes quite significantly with increasing resolution, however. In the tropical lower stratosphere of the 3° model the zonal-mean zonal wind displays a very weak (∼3 m s−1 peak to peak) interannual variation, which-while rather irregular-does display a roughly biennial period and the downward phase propagation that is characteristic of the observed quasi-biennial oscillation.