The Atmospheric Model Intercomparison Project (AMIP) was initiated in 1989 to systematically validate and diagnose the performance of atmospheric general circulation models (GCMs). By 1995, 31 modeling groups had contributed monthly mean data for selected statistics, which were analyzed by participating groups and AMIP diagnostic subprojects. The overall performance of the models in simulating large-scale seasonal distributions of pressure, temperature, and circulation was found to be reasonable, although there were notable systematic errors in cloudiness, precipitation, and ocean surface heat flux. The models generally reproduced the seasonal cycle of sea level pressure and interannual variability of sea level pressure in the tropical Pacific, but had less success in simulating midlatitude interannual variability. A subset of models revisited the experiment with revised versions, showing a reduction in systematic errors in cloudiness but only a slight improvement in other variables. The AMIP project has documented the systematic errors in atmospheric GCMs and served as a reference for model sensitivity and predictability experiments. The continuation of AMIP II aims to address a wider range of variables and processes using an improved diagnostic and experimental infrastructure.The Atmospheric Model Intercomparison Project (AMIP) was initiated in 1989 to systematically validate and diagnose the performance of atmospheric general circulation models (GCMs). By 1995, 31 modeling groups had contributed monthly mean data for selected statistics, which were analyzed by participating groups and AMIP diagnostic subprojects. The overall performance of the models in simulating large-scale seasonal distributions of pressure, temperature, and circulation was found to be reasonable, although there were notable systematic errors in cloudiness, precipitation, and ocean surface heat flux. The models generally reproduced the seasonal cycle of sea level pressure and interannual variability of sea level pressure in the tropical Pacific, but had less success in simulating midlatitude interannual variability. A subset of models revisited the experiment with revised versions, showing a reduction in systematic errors in cloudiness but only a slight improvement in other variables. The AMIP project has documented the systematic errors in atmospheric GCMs and served as a reference for model sensitivity and predictability experiments. The continuation of AMIP II aims to address a wider range of variables and processes using an improved diagnostic and experimental infrastructure.