We studied selection by predators for cryptic prey coloration in a visually heterogeneous habitat that consists of two microhabitats. It has been suggested that the probability of escaping detection in such habitats might be optimized by maximizing crypsis in one of the microhabitats. However, a recent model indicates that a coloration that compromises the requirements of different microhabitats might sometimes be the optimal solution. To experimentally study these hypotheses, we allowed great tits (Parus major L.) to search for artificial prey items in two different microhabitats (background boards): small patterned and large patterned. On each board there was one prey item that was either small-patterned, large-patterned or medium-patterned and thus compromised. Search time was used as the measure of crypsis and was on average longer on the large-patterned than on the small-patterned background. On the small-patterned background, the small-patterned prey was more cryptic than the compromised prey, which was in turn more cryptic than the large-patterned prey. On the large-patterned background, the small-patterned prey was least cryptic, but the compromised prey did not differ significantly from the large-patterned prey. The compromised coloration had lower predation risk than the matching colorations. This indicates that in some conditions a compromised coloration might be the best strategy for the prey and has important implications for the study of animal coloration.