For any given area, diurnal variation of the emitted longwave radiation reveals the response of the surface and atmosphere to the astronomical diurnal forcing. However, the diurnal rhythm can be masked locally by changes in cloud cover linked to passage of weather systems. To eliminate this weather “noise” we construct histograms of the pixel radiance counts from the METEOSAT IR window channel, for regions of dimensions 250 to 1000 km, and we study their evolution with time. We find that considerable information on the diurnal variation, both of surface temperature and of cloud cover at all levels, can be obtained from short sequences (3 days) of hourly METEOSAT data. In West Africa, diurnal variation of clear-sky emergent radiances, corresponding to amplitudes of 30–50 K in surface skin temperature, is found not only in the Saharan desert regions but also during the dry season in the Sahel–Sudan–Guinea zone. In the August wet season, in this last area, the amplitude is lower, while diurnally varying cloud cover is evident at middle and upper levels. We find that the bias in sun-synchronous satellite estimates of cloud cover can he as high as 25–30%, depending on the time of observation. Over Brazil, surface temperature cycling is present but smaller than observed in West Africa. Apparent cloud top height is found to undergo a regular systematic decrease through the late night and morning, although as in other tropical land regions, high-level cloud exhibits an early-evening maximum. Using the histogram diurnal variation to discriminate between low cloud and sea, we find that low cloud over the central and eastern South Atlantic exhibits a very strong diurnal/nocturnal cycle, with maximum cover close to local sunrise. Computing the effect of the diurnal variation on the radiation balance of the area, we find a significant weakening of the cloud albedo erred. If daytime sun-synchronous satellite data alone are used to estimate cloud cover, bias as high as 25% (15–20 W m−2) can result in the calculation of radiation balance.