Feeding behaviour of the malaria vector Anopheles arabiensis Patton (Diptera: Culicidae) was monitored for 12 months (March 2003-February 2004) in the Konso District of southern Ethiopia (5 degrees 15'N, 37 degrees 28'E). More than 45 000 An. arabiensis females were collected by host-baited sampling methods (light-traps, human landing catches, cattle-baited traps) and from resting sites (huts and pit shelters). In the village of Fuchucha, where the ratio of cattle : humans was 0.6 : 1, 51% of outdoor-resting mosquitoes and 66% of those collected indoors had fed on humans, human baits outdoors caught > 2.5 times more mosquitoes than those indoors and the mean catch of mosquitoes from pit shelters was about five times that from huts. Overall, the vast majority of feeding and resting occurred outdoors. In the cattle camps of Konso, where humans slept outdoors close to their cattle, approximately 46% of resting mosquitoes collected outdoors had fed on humans despite the high cattle : human ratio (17 : 1). In both places, relatively high proportions of bloodmeals were mixed cow + human: 22-25% at Fuchucha and 37% in the cattle camps. Anthropophily was also gauged experimentally by comparing the numbers of mosquitoes caught in odour-baited entry traps baited with either human or cattle odour. The human-baited trap caught about five times as many mosquitoes as the cattle-baited one. Notwithstanding the potential pitfalls of using standard sampling devices to analyse mosquito behaviour, the results suggest that the An. arabiensis population is inherently anthropophagic, but this is counterbalanced by exophagic and postprandial exophilic tendencies. Consequently, the population feeds sufficiently on humans to transmit malaria (sporozoite rates: 0.3% for Plasmodium falciparum and 0.5% for P. vivax, by detection of circumsporozoite antigen) but also takes a high proportion of meals from non-human hosts, with 59-91% of resting mosquitoes containing blood from cattle. Hence, classical zooprophylaxis is unlikely to have a significant impact on the malaria vectorial capacity of An. arabiensis in Konso, whereas treating cattle with insecticide might do.