Wild-caught Anopheles gambiae s. l. and An. funestus were dissected and their midguts were examined for the presence of Plasmodium falciparum oocyst infections. The mean intensity of infection and the prevalence of infected mosquitoes were determined for each sample, with one sample representing the mosquitoes caught in a single house at any given time. The patterns of infection were investigated using the relationships between prevalence, intensity, and variance within samples, and were found to be consistent with laboratory infections. The overall distribution of oocysts is characterized by a mixture of negative binomial distributions with means determined by the infectiousness of the human hosts, and a constant degree of aggregation (k = 0.0767) presumably determined by the development of oocysts within mosquitoes. The prevalence/intensity relationship was treated as a bivariate distribution to ascertain the effect of sample size on the accuracy of estimation, and to allow inference of intensity from prevalence. In mathematical models fitted to the collected data, sample size affected directly the minimum possible prevalence of infection, and the accuracy of both mean and prevalence estimations. Based on minimum possible positive prevalence rates, data from samples of less than 20-25 mosquitoes would provide unacceptable errors in prevalence estimations. However, natural oocyst rates are consistently higher than the minimum prevalence, and it is suggested that any interpretations from samples of less than approximately 40 mosquitoes must be treated with some caution. Such variation in natural samples means that prediction of intensity of infection from prevalence (or vice versa) is extremely inaccurate.