OBJECTIVE: To compare the cost-effectiveness of malaria treatment based on presumptive diagnosis with that of malaria treatment based on rapid diagnostic tests (RDTs). METHODS: We calculated direct costs (based on experience from Ethiopia and southern Sudan) and effectiveness (in terms of reduced over-treatment) of a free, decentralised treatment programme using artesunate plus amodiaquine (AS + AQ) or artemether-lumefantrine (ART-LUM) in a Plasmodium falciparum epidemic. Our main cost-effectiveness measure was the incremental cost per false positive treatment averted by RDTs. RESULTS: As malaria prevalence increases, the difference in cost between presumptive and RDT-based treatment rises. The threshold prevalence above which the RDT-based strategy becomes more expensive is 21% in the AS + AQ scenario and 55% in the ART-LUM scenario, but these thresholds increase to 58 and 70%, respectively, if the financing body tolerates an incremental cost of 1 euro per false positive averted. However, even at a high (90%) prevalence of malaria consistent with an epidemic peak, an RDT-based strategy would only cost moderately more than the presumptive strategy: +29.9% in the AS + AQ scenario and +19.4% in the ART-LUM scenario. The treatment comparison is insensitive to the age and pregnancy distribution of febrile cases, but is strongly affected by variation in non-biomedical costs. If their unit price were halved, RDTs would be more cost-effective at a malaria prevalence up to 45% in case of AS + AQ treatment and at a prevalence up to 68% in case of ART-LUM treatment. CONCLUSION: In most epidemic prevalence scenarios, RDTs would considerably reduce over-treatment for only a moderate increase in costs over presumptive diagnosis. A substantial decrease in RDT unit price would greatly increase their cost-effectiveness, and should thus be advocated. A tolerated incremental cost of 1 euro is probably justified given overall public health and financial benefits. The RDTs should be considered for malaria epidemics if logistics and human resources allow.