In his stimulating discussion, Wolfgang Wickler criticizes fuzzy usage of term mimicry by drawing attention to its original definition by H. Bates. Mimicry refers to functional ‘model–mimic–selecting agent’ trinity (with varying number of species involved) when the selecting agent (i.e. signal receiver) responds similarly to mimic and model to the advantage of the mimic. Concurring with Wickler I argue that convergence is neither necessary nor sufficient to support similarity as evidence for mimicry and that it is artificial and unproductive to classify mimicry with respect to ontogeny (innate vs. learned similarity) or model species identity (learning from conspecifics vs. heterospecifics). Using butterfly ‘eye’-spots, I argue that just identifying each of the supposed model, the mimic and the selective agent, and even demonstrating that mimic-model similarity affects the agent's behaviour, provides no conclusive evidence for mimicry. Even a demonstration that the mimic benefits from receiver response may not provide conclusive evidence for mimicry. Using avian brood parasite–host egg and nestling mimicry, I emphasize that without experimental manipulation of the hypothesized mimetic traits, it is impossible to test the mimicry hypothesis robustly. Due to fundamental constraints on human perception, some cases of mimicry may in fact be just a by-product of human inability to perceive relevant differences between animal phenotypes (what is similar for human eye, nose or ear may not be viewed, smelled or heard as similar for relevant animal observers), whereas many cases of real mimicry may escape our attention from the same reason (‘hidden’ mimicry). Surprisingly, the same mimetic phenotype may show completely different effects on selective agents under different ecological circumstances. Finally, relatively dissimilar species may be more mimetic than highly similar model–mimic pairs because mimicry may be more fruitfully understood as a co-evolutionary process rather than a similarity.