The advent of genetically modified organisms such as pathogens has raised ecological questions that need to be addressed in order to assess any risks involved in their use. The baculovirus Autographa californica nucleopolyhedrovirus (AcNPV), which infects a number of lepidopteran species, has been modified to express an insect-selective toxin. This genetic modification increases the speed with which it kills its host. However, in addition to this intended feature of the modified virus, there may be other consequences for the host–pathogen interaction. We report a field experiment in which transmission patterns of the wild-type and the genetically modified baculovirus are measured within and between a model target (susceptible) and nontarget (less susceptible) lepidopteran species. Two foliar feeders were chosen: Trichoplusia ni, the cabbage looper, is highly susceptible to this pathogen, while Mamestra brassicae, the cabbage moth, is semipermissive. These two species are used as both the source and the recipients of infection for both virus types. A series of models are fitted to determine the probabilities of infection (given survival from other sources of mortality) over a 7-d period within contained field cages. Fitting these models to data illustrates that a substantial fraction of the population escapes infection, and it is the size of the pathogen-free refuge that varies between treatments. When infected individuals from the less susceptible species die, the yield of virus is greater than from susceptible hosts, yet this does not significantly alter the risk of transmission to other hosts. In contrast, the genetically modified baculovirus always results in a lower risk of infection in the field compared to the wild type. This is because the recombinant virus causes paralysis, and as a result, the cadaver may fall from the plant before death and virus release. Hence the number of cadavers remaining on the foliage has a greater influence on transmission than the yield of virus from those cadavers.