Water vapor simulations from the Atmospheric Model Intercomparison Project have been compared with satellite observations of precipitable water and upper tropospheric humidity. The model means and standard deviations of precipitable water compare favorably with observations and most models also reproduce the observed seasonal cycle well. The models show a regional dependence in reproducing the observed interannual variability of precipitable water, doing better in the deep tropics than in the subtropics and midlatitudes. The skill in reproducing the observed interannual variability in the deep tropics is attributed to accurate reproduction of anomalies during the 1982--1983 El Ni¿o-Southern Oscillation event. Comparisons between modeled and observed upper tropospheric humidity are more problematic, but significant differences are found in both the mean values and the ability of the models to produce the observed seasonal and interannual variability. Additional comparisons of both models and observations are required to resolve these discrepancies. Examination of leading modes of interannual variability shows that models that do a better job in simulating the observed all-sky outgoing longwave radiation in the tropics also do a better job of simulating the observed upper tropospheric humidity.¿ 1997 American Geophysical Union