(A) Properties of the experimental prey communities used in this study, with examples. All communities had a 1:1 ratio of ‘good’ prey to ‘bad’ prey. A reliable trait allowed perfect discrimination. The richness and evenness of its values varied between experiments. An unreliable trait that did not vary between experiments yielded less accurate discrimination. The exact distribution of prey in each community is given below its richness and evenness statistics, with numbers to indicate the abundance of each prey. As drawn here, shape is the reliable trait (e.g. circle = good, star = bad), whereas color is the unreliable trait (blue = good 78% of the time, yellow = bad 78% of the time). Red boxes indicate the focal prey that were compared across experiments in panels B and C (their actual colors and shapes differed among treatments). (B) Total discrimination subjects exhibited towards focal prey, that is the summed influence of both reliable and unreliable traits. (C) Subjects’ relative use of the reliable trait compared with the unreliable trait for discrimination, that is the difference between the effect of reliable and unreliable traits. The y-axis indicates the difference in the ability of the reliable trait to predict behavior compared to the unreliable trait. In (B) and (C), estimates are grouped using the Bonferroni correction for multiple pairwise comparisons, and 95% confidence intervals are shown. See Methods for details on interpreting log-odds.