People can accurately assess the "mood of a crowd" by rapidly extracting the average intensity of all the individual expressions, when the crowd consists of a set of faces comprising different expressions of the same individual. Here, we investigate the processes involved when people judge the expression intensity of individual faces that appear in the context of a more naturalistic crowd of different individuals' faces. We show that judgments of the intensity of happy and angry expressions for individual faces are biased toward the group mean expression intensity, even when the faces are all different individuals. In a second experiment, we demonstrate that this bias is not due to a generic tendency to endorse intermediate intensity expressions more frequently than more extreme intensity expressions. Together, these findings suggest that people integrate ensemble information about the group average expression when they make judgments of individual faces' expressions.