Confidence limits of expected twenty-day rainfall have been computed for two widely separated maize-producing regions in Kenya. The 35-km. square survey areas were quite different topographically, and limits were derived for seven rainfall sites in each. Ratios of these limits to the corresponding long-term mean precipitations were evaluated for each site for each twenty-day period during the appropriate maize-growing seasons. For a particular period and confidence level, these ratios were almost constant for either group of seven sites. The variability computed for a single, well-chosen site is therefore representative of quite a large neighbourhood. Expected rainfall totals were also derived for a ‘fictitious’ site, simulating the average precipitation over the whole of the flatter of the two areas. The resulting limits were less extreme than those of any individual site, indicating that areal rainfall, as used in hydrology, is less variable than the agriculturally-important point rainfall.