The purpose of this paper is to determine the optimal space-time resolution which is suitable for the estimation of cumulative precipitation by a simple rainfall-radiance relation. The correlations between precipitation ground measurements and convective clouds derived from Meteosat infrared brightness temperatures are computed for different spatial and temporal scales, over the tropical Sahelian region and over a validation site near Niamey (Republic of Niger). It is shown that monthly precipitations display a higher spatial variability than the monthly cold cloud field, causing a low correlation between them. The data set obtained during August 1988 from the dense network of recording raingauges of theEPSAT-NIGERexperiment is used for a study at short time scale. Several infrared cloud indices are compared and it is found that the mean precipitation estimated on a large surface (> 100 by 100km2) is well enough correlated to the satellite infrared index cumulated on daily or even shorter time periods, so that about 70 per cent of the variance of rainfall estimated from these radiances is explained by a linear function of the fraction of surface covered by cold clouds.