Predation risk is allegedly reduced in Batesian and Müllerian mimics, because their coloration resembles the conspicuous coloration of unpalatable prey. The efficacy of mimicry is thought to be affected by variation in the unpalatability of prey, the conspicuousness of the signals, and the visual system of predators that see them. Many frog species exhibit small colorful patches contrasting against an otherwise dark body. By measuring toxicity and color reflectance in a geographically variable frog species and the syntopic toxic species, we tested whether unpalatability was correlated with between-species color resemblance and whether resemblance was highest for the most conspicuous components of coloration pattern. Heterospecific resemblance in colorful patches was highest between species at the same locality but unrelated to concomitant variation in toxicity. Surprisingly, resemblance was lower for the conspicuous femoral patches compared to the inconspicuous dorsum. By building visual models, we further tested whether resemblance was affected by the visual system of model predators. As predicted, mimic-model resemblance was higher under the visual system of simulated predators compared to no visual system at all. Our results indicate that femoral patches are aposematic signals and support a role of mimicry in driving phenotypic divergence or mimetic radiation between localities.