Müllerian mimicry, where unpalatable prey share common warning patterns, has long fascinated evolutionary biologists. It is commonly assumed that Müllerian mimics benefit by sharing the costs of predator education, thus reducing per capita mortality, although there has been no direct test of this assumption. Here, we specifically measure the selection pressure exerted by avian predators on unpalatable prey with different degrees of visual similarity in their warning patterns. Using wild–caught birds foraging on novel patterned prey in the laboratory, we unexpectedly found that pattern similarity did not increase the speed of avoidance learning, and even dissimilar mimics shared the education of naive predators. This was a consistent finding across two different densities of unpalatable prey, although mortalities were lower at the higher density as expected. Interestingly, the mortalities of Müllerian mimics were affected by pattern similarity in the predicted way by the end of our experiment, although the result was not quite significant. This suggests that the benefits to Müllerian mimics may emerge only later in the learning process, and that predator experience of the patterns may affect the degree to which pattern similarity is important. This highlights the need to measure of real predators if we are to understand fully the evolution of mimicry systems.