Aposematic prey advertise their unprofitability with conspicuous warning signals that are often composed of multiple color patterns. Many species show intraspecific variation in these patterns even though selection is expected to favor invariable warning signals that enhance predator learning. However, if predators acquire avoidance to specific signal components, this might relax selection on other aposematic traits and explain variability. Here, we investigated this idea in the aposematic moth Amata nigriceps that has conspicuous black and orange coloration. The size of the orange spots in the wings is highly variable between individuals, whereas the number and width of orange abdominal stripes remains consistent. We produced artificial moths that varied in the proportion of orange in the wings or the presence of abdominal stripes. We presented these to a natural avian predator, the noisy miner (Manorina melanocephala), and recorded how different warning signal components influenced their attack decisions. When moth models had orange stripes on the abdomen, birds did not discriminate between different wing signals. However, when the stripes on the abdomen were removed, birds chose the model with smaller wing spots. In addition, we found that birds were more likely to attack moths with a smaller number of abdominal stripes. Together, our results suggest that bird predators primarily pay attention to the abdominal stripes of A. nigriceps, and this could relax selection on wing coloration. Our study highlights the importance of considering individual warning signal components if we are to understand how predation shapes selection on prey warning coloration.