OBJECTIVES: The duration of untreated Plasmodium falciparum infections in naturally exposed human populations is of interest for rational planning of malaria control interventions as it is related to the duration of infectivity. The extent of variability in duration is relevant where transmission is seasonal, and for the planning of elimination efforts. Methods for measuring these quantities from genotyping data have been restricted to exponential models of infection survival, as implied by constant clearance rates. Such models have greatly improved the understanding of infection dynamics on a population level but likely misrepresent the within-host dynamics of many pathogens. Conversely, the statistical properties of the distribution of infection durations, and how these are affected by exposure, should contain information on within-host dynamics. METHODS AND RESULTS: We extended existing methods for the analysis of longitudinal genotyping data on P. falciparum infections. Our method simultaneously estimates force of infection, detectability, and the distribution of infection durations. Infection durations are modeled using parametric survival distributions. The method is validated using simulated data, and applied to data from a cohort study in Navrongo, Northern Ghana. Distribution estimates from exponential, Weibull, lognormal, and gamma models are compared with the distribution of durations in malariatherapy data. CONCLUSIONS: The Weibull model fitted the data best. It estimated a shorter mean duration than the exponential model, which gave the worst fit. The distribution estimates appeared positively skewed when compared with the distribution of durations in malariatherapy data, suggesting that a significant proportion of infections is cleared shortly after inoculation. We conclude that malariatherapy data, the most important source of information on P. falciparum within-host dynamics, may not be representative of the actual processes in natural populations, and should be used with care. Further, conclusions from transmission models assuming exponential infection survival may be biased.