The methods used for low-level aerial census in Africa, which have been developed by wildlife biologists, were adapted to suit an agricultural research team counting pastoral livestock. The study area in Kenyan Maasailand was small by aerial survey standards (3510 km 2). It was further sub-divided into a northern (870 km 2) and southern (2640 km 2) census zone because of differences in ecology and stage of development between the Group Ranches in the north and south. The design of each survey was a compromise between the need to count nomadic livestock, which is best done by stratified random sampling, and the need to relate spatial and temporal variations in animal density to the natural resources of the area using unstratified systematic sampling. A basic pattern of systematic reconnaissance flights was adopted with flight lines aligned with the UTM grid, to maintain compatibility with surveys by other scientists, but in this case at a closer transect spacing of 2·5 km. Stratified systematic sampling was then applied to the southern census zone when there was prior knowledge, either from ground truth or preliminary aerial reconnaissance, of areas devoid of livestock. The flying time was then concentrated over the strata or sample zones containing livestock. The northern zone was too small to stratify except to the extent of total counting livestock at watering points. Both forms of stratification improved the precision of the population estimate compared with that obtained by unstratified systematic sampling. Further improvements in dry season counts were obtained by delaying take-off from the morning, when livestock were clumped around watering points, to the afternoon when they had dispersed to their grazing areas. Each survey was modified in the light of the experience of the previous one, and a trend of increasing precision in the livestock population estimates was achieved. There was more scope for improvement in the southern census zone where the standard error dropped from 25 to 12% in cattle and 47 to 13% in smallstock within four surveys. Best estimates of the livestock populations in both the census zones had an average standard error of 10% which compared favourably with other surveys with similar numbers of sample units ( x 17) and sample fractions ( x 20%). The estimates of the livestock population obtained by aerial census were compared with ground counts of the same area. In the northern census zone, the figures from the ground counts were considered more reliable except on Merueshi where they failed to account for at least 2000 uninvited cattle. In the more rugged terrain and larger area of the southern census zone, aerial census was found to be more precise and efficient than counting on the ground. There is a growing awareness of the need for a high standard of reliability in aerial data which is even more relevant to agricultural than wildlife research. Not only are agricultural field workers better placed to check the figures on the ground, but they are more likely to translate the results into management.