We test the ability of ten atmospheric General Circulation Models (GCMs) participating in AMIP to reproduce the observed seasonal and interannualvariability of precipitable water (W) and the clear-sky greenhouse effect (G). Using satellite data (SSM-I, ISCCP/TOVS and ERBE), we first develop simplestatistical analyses to quantify the rate of variation of W and G as a function of the local variation in the sea surface temperature (SST). The same analysisprocedure is then applied to the GCM output in order to evaluate their representation of these seasonal and interannual sensitivities. This evaluation revealsthat certain biases are common to most models. For example, most GCMs tend to underestimate mean values of both W and G in the tropics compared toobservations. Also in the tropics, the modelled interannual (seasonal) sensitivity of the clear-sky greenhouse effect to a local SST change tends to be toolarge (small). However, there are also differences between the models themselves that are not clearly related to the various parametizations. To trace thesource of these differences, we need more information about the variability of the vertical structure (temperature and water vapor) of the atmosphere. Theinfluence of the variability of the atmospheric vertical structure is investigated using ECMWF meteorological analyses. The results are used to analyse the origin of the seasonal variation of precipitable water and the clear-sky greenhouse effect for a few models.