The "geographic mosaic" approach to understanding coevolution is predicated on the existence of variable selection across the landscape of an interaction between species. A range of ecological factors, from differences in resource availability to differences in community composition, can generate such a mosaic of selection among populations, and thereby differences in the strength of coevolution. The result is a mixture of hotspots, where reciprocal selection is strong, and coldspots, where reciprocal selection is weak or absent, throughout the ranges of species. Population subdivision further provides the opportunity for nonadaptive forces, including gene flow, drift, and metapopulation dynamics, to influence the coevolutionary interaction between species. Some predicted results of this geographic mosaic of coevolution include maladapted or mismatched phenotypes, maintenance of high levels of polymorphism, and prevention of stable equilibrium trait combinations. To evaluate the potential for the geographic mosaic to influence predator-prey coevolution, we investigated the geographic pattern of genetically determined TTX resistance in the garter snake Thamnophis sirtalis over much of the range of its ecological interaction with toxic newts of genus Taricha. We assayed TTX resistance in over 2900 garter snakes representing 333 families from 40 populations throughout western North America. Our results provide dramatic evidence that geographic structure is an important component in coevolutionary interactions between predators and prey. Resistance levels vary substantially (over three orders of magnitude) among populations and over short distances. The spatial array of variation is consistent with two areas of intense evolutionary response by predators ("hotspots") surrounded by clines of decreasing resistance. Some general predictions of the geographic mosaic process are supported, including clinal variation in phenotypes, polymorphism in some populations, and divergent outcomes of the interaction between predator and prey. Conversely, our data provide little support for one of the major predictions, mismatched values of interacting traits. Two lines of evidence suggest selection is paramount in determining population variation in resistance. First, phylogenetic information indicates that two hotspots of TTX resistance have evolved independently. Second, in the one region that TTX levels in prey have been quantified, resistance and toxicity levels match almost perfectly over a wide phenotypic and geographic range. However, these results do not preclude the role the nonadaptive forces in generating the overall geographic mosaic of TTX resistance. Much work remains to fill in the geographic pattern of variation among prey populations and, just as importantly, to explore the variation in the ecology of the interaction that occurs within populations.