<p>Face ensemble encoding involves synthesizing summary information from groups of faces, providing a mechanism to overcome limitations in visual working memory. Yet, research on the role of attention has revealed mixed findings. Also, the simultaneous processing of summary representations and individual faces within an ensemble remains largely unexplored. Here, across three experiments, participants viewed ensembles without a central face (<i>n</i> = 32), or with a central face, while attention was distributed across the entire ensemble (<i>n</i> = 38) or focused centrally (<i>n</i> = 38). Critically, the consistency of center and surround faces varied as a function of emotional valence (i.e., same versus opposite). Participants completed an expression similarity-rating task between an ensemble and a single face, which was used to recover, via image reconstruction, visual estimates of summary representations. Reconstructions were then assessed against central faces, surrounding faces, and their averages. We show that focused attention enhances central face representation and that consistency benefits the representation of both the center and of the surround. However, central faces outweigh the overall surround representation only when attention is focused on a center face inconsistent with its surround. These findings reveal a flexible relationship between attention and stimulus structure in ensemble perception.</p>