The hypothesis of evasive mimicry proposes that: (a) the conspicuous coloration of some animals has evolved to warn predators that they are good at avoiding capture; and (b) this conspicuous coloration is mimicked by other, less evasive species which thus gain protection from predation. We have tested these ideas with wild birds as the predators and red and yellow pastry ‘baits’ as the prey. We used a specially constructed bird table to make one of the colours good at ‘escaping’: when a bird was about to eat a bait the observer pulled a cord which activated a mechanism that made the baits of one colour (but not the other) drop out of the bird's reach. We first measured the overall selection by the naive birds by presenting, for one day, a ‘control’ of equal numbers of non‐evasive red and non‐evasive yellow baits. The birds were then trained for 20 days on equal numbers of evasive yellow and non‐evasive red baits, and were then again given a second control (in which the yellows were now non‐evasive ‘mimics’). The training was repeated for 18 days but with red as the evasive colour, and a third control was then presented (with reds now as the mimics). The proportions attacked daily in the two training sessions suggested that the birds were learning to ignore yellows in the first and reds in the second. In comparison with the first control, fewer yellows were eaten after they had been trained to ignore escaping yellows, and more were eaten after they had been trained on escaping reds. We conclude that the results are further evidence for the hypoth