Persistence is a central issue in population ecology with important implications for population management. Most theoretical studies have focused on continually interacting populations, even though many systems are subject to ecological disturbances which confound analysis of persistence. In this paper, we use a combination of a simple parasite–hyperparasite model with disturbances and field data to investigate the factors contributing to the observed persistence of the parasite population. The field data are taken from a two–year experiment (including five growing seasons) investigating the use of the mycoparasite Sporidesmium sclerotivorum as a persistent biological control agent of Sclerotinia minor, an economically important fungal parasite of lettuce. We show that the standard assumption of homogeneous mixing fails to predict the observed persistence of the parasite population. We demonstrate that allowing for heterogeneous mixing prevents the fade–out predicted in the homogeneous mixing case. The implications of the results for broad classes of host–parasite systems are discussed.