Understanding the early evolution of aposematic (warning) coloration has been a challenge for scientists, as a new conspicuous morph in a population of cryptic insects would have a high predation risk and would probably die out before local predators learnt to avoid it. Fisher presented the idea of aggregation benefit through the survival of related individuals; however, his theory has been strongly debated as the mechanisms that favour grouping have never been explored experimentally with the incorporation of detectability costs. Here we create a comprehensive 'novel world' experiment with the great tit (Parus major) as a predator to explore simultaneously the predation-related benefits and costs for aposematic aggregated prey, manipulating both group size and signal strength. Our results show that grouping would have been highly beneficial for the first aposematic prey individuals surrounded by naive predators, because (1) detectability risk increased only asymptotically with group size; (2) additional detectability costs due to conspicuous signals were marginal in groups; (3) even naive predators deserted the group after detecting unpalatability (dilution effect); and (4) avoidance learning of signal was faster in groups. None of these mechanisms require kin selection.