Outbreaks of forest-defoliating insects can have severe impacts on forests, exacerbating climate change, but outbreaks would be even worse if not for epizootics of fatal, species-specific pathogens. An understanding of what determines the timing and severity of pathogen epizootics in forest insects could make it easier to mitigate the effects of outbreaks, but achieving such an understanding is a difficult problem. Work in my lab attempts to solve this problem by using data to choose between competing models of disease dynamics. This approach is common to much of disease ecology, in contrast to many diseases, insect pathogens can be easily used in experiments. It is therefore straightforward, to directly test mechanistic models of disease transmission. By using Bayesian statistical approaches to combine experimental and observational data, we have been able to show that small-scale transmission mechanisms often play a key role in driving large-scale epizootics. This approach has allowed us to disentangle the effects of weather and host density on the spread of a fungal pathogen (Entomophaga maimaiga) of the gypsy moth (Lymantria dispar), which in turn has allowed us to show that the ability of the pathogen to control the insect will likely decline in the future because of global warming. Similarly, we have shown that host heterogeneity in infection risk interacts with host density to determine the severity of epizootics of a baculovirus pathogen of the Douglas-fir tussock moth. We are therefore using our models to guide efforts by the USDA Forest Service to use the baculovirus as an environmentally benign insecticide during tussock moth outbreaks. Combining experimental data with general models of disease spread can thus provide significant assistance to insect pest-control efforts.