Predators typically avoid dangerous species, and batesian mimicry evolves when a palatable species (the 'mimic') co-opts a warning signal from a dangerous species (the 'model') and thereby deceives its potential predators. Because predators would not be under selection to avoid the model and any of its look-alikes in areas where the model is absent (that is, allopatry), batesian mimics should occur only in sympatry with their model. However, contrary to this expectation, batesian mimics often occur in allopatry. Here we focus on one such example--a coral snake mimic. Using indirect DNA-based methods, we provide evidence suggesting that mimics migrate from sympatry, where mimicry is favoured, to allopatry, where it is disfavoured. Such gene flow is much stronger in nuclear genes than in maternally inherited mitochondrial genes, indicating that dispersal by males may explain the presence of mimetic phenotypes in allopatry. Despite this gene flow, however, individuals from allopatry resemble the model less than do individuals from sympatry. We show that this breakdown of mimicry probably reflects predator-mediated selection acting against individuals expressing the more conspicuous mimetic phenotype in allopatry. Thus, although gene flow may explain why batesian mimics occur in allopatry, natural selection may often override such gene flow and promote the evolution of non-mimetic phenotypes in such areas.