1. Although background matching decreases prey detectability, resemblance between camouflaged prey and their visual background is seldom perfect. This could be because even a moderate resemblance might provide sufficient protection, and additional adjustment of colour pattern might give little benefit. Alternatively, close resemblance to background may not be attained due to trade-offs or constraints. To understand selection on colour patterns of camouflaged prey and the existence of inaccurate background matching, it is necessary to investigate how detectability of a colour pattern varies with its resemblance to the background. 2. We trained wild-caught blue tits (Cyanistes caeruleus) to search for artificial prey. We manipulated the resemblance of the artificial prey items to the visual backgrounds. 3. For the first half of the twelve repeated prey presentations we found a non-linear relationship between resemblance and detectability, such that for prey that had high background matching, a change in resemblance resulted in a larger change in detectability than an equal change in resemblance did for prey with lower background matching. However, for the second half of the presentations this relationship was linear. Moreover, in a two-patch-type habitat a prey pattern that was a compromise between the two different backgrounds did after few initial presentations equally well as the prey pattern that matched highly one of the backgrounds. 4. Our results indicate an intense selection for close matching in a single background. Yet, in the heterogeneous environment that consisted of two backgrounds the compromise, which only loosely resembled either background, provided good protection. Therefore, we conclude that cryptic colour patterns that bear only a loose resemblance to a given background, and thus represent inaccurate background matching, may be adaptive outcomes.