Faces are not processed holistically in ensemble judgments

Holistic processing is a hallmark of individual face perception, but whether it extends to ensemble judgments of faces remains unknown. We tested two operationalizations of holistic processing: (1) failures of selective attention to face parts, and (2) sensitivity to configural arrangement. In Experiment 1, participants judged the average identity of top-halves in face ensembles where the most prevalent top-half was paired with homogeneous versus diverse bottom-halves. Bayesian evidence favoured no difference between conditions, indicating bottom-halves could be selectively ignored. Experiment 2 replicated this finding comparing ensembles with versus without bottom-halves present. In contrast, Experiment 3 showed that when participants judged whole-face averages, both the number of face parts and their correct configuration influenced accuracy, though configural effects were smaller than part-based effects. These results demonstrate that ensemble face perception does not produce the failures of selective attention characteristic of individual face recognition, while configural information does contribute when task-relevant. This dissociation between two effects often attributed to the same construct supports theoretical accounts where holistic processing arises from learned attentional weights that are context-dependent rather than from obligatory holistic representations.