Conspicuous warning coloration helps to protect prey because it signals to potential predators that the prey is unprofitable. However, such signals only work once predators have come to associate the conspicuous colour with the unprofitability of the prey. The evolution of warning coloration is generally considered to be paradoxical, because it has traditionally been assumed that the first brightly coloured individuals would be at an immediate selective disadvantage because of their greater conspicuousness to predators that are naïve to the meaning of the signal. As a result, it has been difficult to understand how a novel conspicuous colour morph could ever avoid rapid extinction, and instead survive and spread in the population until predators have become educated about the signal. In the present study, we experimentally simulated the appearance of a single novel coloured mutant in small populations (20 individuals) of palatable artificial pastry “prey”. The colour morph frequencies in each “generation” of prey (presented on successive days of a trial) were determined by the relative survival of the previous generation under predation by free-living birds. We found that the novel colour morphs regularly persisted and increased from a starting frequency of 1/20 to reach fixation (100%), despite being fully palatable, even when the novel morph was much more conspicuous against the background than the familiar morph. This was true for both green (not normally considered a warning colour) and red (a classic warning colour) novel morphs. Novel colours reached fixation significantly faster than could be accounted for by random drift, indicating differential predation in relation to prey colour by the birds. Our experiments show that the immediate demise of a fully palatable new prey morph is not an inevitable outcome of predator behaviour, because even very conspicuous prey can gain protection from conservative foragers, simply by being novel.