Many species defend themselves against enemies using repellent chemicals. An important but unanswered question is why investment in chemical defence is often variable within prey populations. One explanation is that some prey benefit by cheating, paying no costs of defence, but gaining a reduced attack rate because of the presence of defended conspecifics. Two important assumptions about predator behaviour must be met to explain cheating as a stable strategy: first, predators increase attack rates as cheats increase in frequency; second, defended prey survive attacks better than non-defended conspecifics. We lack data from wild predators that evaluate these hypotheses. Here, we examine how changes in the frequency of non-defended ‘cheats’ affect predation by wild birds on a group of otherwise defended prey. We presented mealworm larvae that were either edible (‘cheats’) or unpalatable (bitter tasting), and varied the proportion of cheats from 0 to 1 by increments of 0.25. We found strong frequency-dependent effects on the birds' foraging behaviour, with the proportion of prey attacked increasing nonlinearly with the frequency of cheats. We did not, however, observe that birds taste-rejected defended prey at the site of capture. One explanation is that wild birds may not assess prey palatability at the site of capture, but do this elsewhere. If so, defended and undefended prey may pay high costs of initial attack and relocation away from ecologically favourable locations. Alternatively, defended prey may not be taste-rejected because with acute time constraints, wild birds do not have time to make fine-grained decisions during feeding. We discuss the data in relation to the evolutionary ecology of prey defences.