We present an analysis of the seasonal precipitation associated with the African, Indian and the Australian-Indonesian monsoon and the interannual variation of the Indian monsoon simulated by 30 atmospheric general circulation models undertaken as a special diagnostic subproject of the Atmospheric Model Intercomparison Project (AMIP). The seasonal migration of the major rainbelt observed over the African region, is reasonably well simulated by almost all the models. The Asia West Pacific region is more complex because of the presence of warm oceans equatorward of heated continents. Whereas some models simulate the observed seasonal migration of the primary rainbelt, in several others this rainbelt remains over the equatorial oceans in all seasons. Thus, the models fall into two distinct classes on the basis of the seasonal variation of the major rainbelt over the Asia West Pacific sector, the first (class I) are models with a realistic simulation of the seasonal migration and the major rainbelt over the continent in the boreal summer; and the second (class II) are models with a smaller amplitude of seasonal migration than observed. The mean rainfall pattern over the Indian region for July-August (the peak monsoon months) is even more complex because, in addition to the primary rainbelt over the Indian monsoon zone (the monsoon rainbelt) and the secondary one over the equatorial Indian ocean, another zone with significant rainfall occurs over the foothills of Himalayas just north of the monsoon zone. Eleven models simulate the monsoon rainbelt reasonably realistically. Of these, in the simulations of five belonging to class I, the monsoon rainbelt over India in the summer is a manifestation of the seasonal migration of the planetary scale system. However in those belonging to class II it is associated with a more localised system. In several models, the oceanic rainbelt dominates the continental one. On the whole, the skill in simulation of excess/deficit summer monsoon rainfall over the Indian region is found to be much larger for models of class I than II, particularly for the ENSO associated seasons. Thus, the classification based on seasonal mean patterns is found to be useful for interpreting the simulation of interannual variation. The mean rainfall pattern of models of class I is closer to the observed and has a higher pattern correlation coefficient than that of class II. This supports Sperber and Palmer's (1996) result of the association of better simulation of interannual variability with better simulation of the mean rainfall pattern. The hypothesis, that the skill of simulation of the interannual variation of the all-India monsoon rainfall in association with ENSO depends upon the skill of simulation of the seasonal variation over the Asia West Pacific sector, is supported by a case in which we have two versions of the model where NCEP1 is in class II and NCEP2 is in class I. The simulation of the interannual variation of the local response over the central Pacific as well as the all-India monsoon rainfall are good for NCEP2 and poor for NCEP1. Our results suggest that when the model climatology is reasonably close to observations, to achieve a realistic simulation of the interannual variation of all-India monsoon rainfall associated with ENSO, the focus should be on improvement of the simulation of the seasonal variation over the Asia West Pacific sector rather than further improvement of the simulation of the mean rainfall pattern over the Indian region.