Bacillus thuringiensis (Bt) and its insecticidal toxins are widely exploited in microbial biopesticides and genetically modified crops. Its population biology is, however, poorly understood. Important issues for the safe, sustainable exploitation of Bt include understanding how selection maintains expression of insecticidal toxins in nature, whether entomopathogenic Bt is ecologically distinct from related human pathogens in the Bacillus cereus group, and how the use of microbial pesticides alters natural bacterial populations. We addressed these questions with a MLST scheme applied to a field experiment in which we excluded/added insect hosts and microbial pesticides in a factorial design. The presence of insects increased the density of Bt/B. cereus in the soil and the proportion of strains expressing insecticidal toxins. We found a near-epidemic population structure dominated by a single entomopathogenic genotype (ST8) in sprayed and unsprayed enclosures. Biopesticidal ST8 proliferated in hosts after spraying but was also found naturally associated with leaves more than any other genotype. In an independent experiment several ST8 isolates proved better than a range of non-pathogenic STs at endophytic and epiphytic colonization of seedlings from soil. This is the first experimental demonstration of Bt behaving as a specialized insect pathogen in the field. These data provide a basis for understanding both Bt ecology and the influence of anthropogenic factors on Bt populations. This natural population of Bt showed habitat associations and a population structure that differed markedly from previous MLST studies of less ecologically coherent B. cereus sample collections. The host-specific adaptations of ST8, its close association with its toxin plasmid and its high prevalence within its clade are analogous to the biology of Bacillus anthracis. This prevalence also suggests that selection for resistance to the insecticidal toxins of ST8 will have been stronger than for other toxin classes.