We reviewed the use of simple mathematical models to estimate the duration of Plasmodium falciparum infection after transmission has been interrupted. We then fit an exponential decay model to repeated cross-sectional survey data collected from three historical trials of indoor residual spraying against malaria: one from two contiguous districts in Tanzania-Kenya (Pare Taveta) carried out in 1954, the others in West Papua (1953), and the Garki project in northern Nigeria (1972-1973). A cross-sectional analysis of these datasets gave overall estimates of 602 days (95% confidence interval [CI] = 581-625) for the infection duration in Pare Taveta, 734 days (95% CI = 645-849) in West Papua, and 1,329 days (95% CI =1,193-1,499) for Garki. These estimates are much greater than the most widely quoted figures for the duration of untreated P. falciparum infections. Although these may be exaggerated because some reinfections occurred despite intensive vector control, prevalence was still decreasing when all these projects ended. Longitudinal survival analysis of the Garki data gave much shorter estimates of duration (186 days, 95% CI = 181-191), but effects of imperfect detection of parasites by microscopy severely bias these estimates. Estimates of infection duration for different age groups showed considerable variation but no general age trend. There was also no clear relationship between malaria endemicity and infection duration. Analyses of successive sampling from the same individuals with parasite typing are needed to obtain more reliable estimates of infection duration in endemic areas. Periods of several years may be required to evaluate long-term effects of interventions on malaria prevalence.