Abstract
Mimicry is considered a classic example of the elaborate adaptations that natural selection can produce, yet often similarity between Batesian (harmless) mimics and their unpalatable models is far from perfect. Variation in mimetic accuracy is a puzzle, as natural selection should favour mimics that are hardest to distinguish from their models. Numerous hypotheses exist to explain the persistence of inaccurate mimics, but most have rarely or never been tested against empirical observations from wild populations. One reason for this is the difficulty in measuring pattern similarity, a key aspect of mimicry. Here, we use a recently developed method, based on the distance transform of binary images, to quantify pattern similarity both within and among species for a group of hoverflies and their hymenopteran models. This allowed us to test three key hypotheses regarding inaccurate mimicry. Firstly, we tested the prediction that selection should be more relaxed in less accurate mimics, but found that levels of phenotypic variation are similar across most hoverfly species. Secondly, we found no evidence that mimics have to compromise between accuracy to multiple model species. However, we did find that darker-coloured hoverflies are less accurate mimics, which could lead to a trade-off between mimicry and thermoregulation in temperate regions. Our results shed light on a classic problem concerning the limitations of natural selection.