OBJECTIVES: Malaria clinical trials need precise endpoints to measure efficacy. In endemic areas where asymptomatic parasitaemia is common, 'fever plus parasitaemia' may not differentiate between malaria cases and non-cases. Case definitions based on parasite cut-off densities may be more appropriate but may vary with age and transmission intensity. This study examines appropriate case definitions from parasitological surveys conducted over a broad range of transmission intensities, using altitude as a proxy for transmission intensity. METHODS: Cross-sectional data collected from 24 villages at different altitudes in an endemic area of northeastern Tanzania were used to calculate malaria-attributable fractions using a modified Poisson regression method. We modelled fever as a function of parasite density and determined the optimum cut-off densities of parasites to cause fever using sensitivity and specificity analyses. RESULTS: The optimum cut-off density varied by altitude in children aged under 5 years: a case definition of 4,000 parasites per mul at altitudes <600 m (high transmission intensity) was most appropriate, compared with 1,000 parasites per mul at altitudes >600 m (low transmission intensity). In children aged over 5 years and adults, there was little variation by altitude and a case definition of any parasites plus fever was the most appropriate. CONCLUSIONS: Locally appropriate case definitions of malaria should be used for research purposes. In our setting, these varied independently with age and transmission intensity.