There is a consensus that malaria is a growing problem in African highlands. This is surprising because many parts of the highlands were considered too cold to support transmission. In this report, we examined how transmission of Plasmodium falciparum in six villages changed along an altitude transect in the Usambara Mountains, Tanzania, from 300 m to 1700 m. Routine entomological collections were made using spray catches and light traps for 15 mo. Direct estimates of entomological inoculation rates and indirect estimates of vectorial capacity suggested a >1000-fold reduction in transmission intensity between the holoendemic lowland and the hypoendemic highland plateau. Lowland transmission was perennial with a significant peak in the cool season after the long rains in May, when vectors densities were high. In the highlands, low temperatures prevented parasite development in mosquitoes during the cool season rains, and highland transmission was therefore limited to the warm dry season when vector densities were low. The primary effect of increasing altitude was a log-linear reduction in vector abundance and, to a lesser extent, a reduction in the proportion of infective mosquitoes. Highland malaria transmission was maintained at extraordinarily low vector densities. We discuss herein the implications of these findings for modeling malaria and suggest that process-based models of malaria transmission risk should be improved by considering the direct effect of temperature on vector densities. Our findings suggest that variation in the short rains in November and changes in agricultural practices are likely to be important generators of epidemics in the Usambaras.