To establish a simple definition of a malaria attack based on blood parasite density and other explanatory covariates, a cohort study was conducted from 1993 to 1996 in the Madagascar highlands undergoing a low seasonal transmission of falciparum malaria. Using logistic regression, the explanatory variables found to be significantly related to the risk of fever are parasite density, age, season, and year. However, and in contrast with other studies, we found no evidence of a clear cutoff in parasite density values consistent with the concept of "pyrogenic threshold" despite a gradual increase of the risk of fever with increasing parasite density. Furthermore, the model evidenced an individual-dependent relationship at a given age. This point was in accordance with the immunological data recorded from the participants. The investigators conclude that the parasite density to distinguish malaria attacks from other causes of fever is not reliable in a context of low falciparum transmission.