Voluminous research supports holistic processing of faces. However, little is known about how holistic processing affects recognition of newly learned faces, a question of importance for improving performance in police lineups and other real-world tasks. Drawing on cognitive and neuropsychological research, we suggest that holistic processing facilitates the formation of unitized representations that support discrimination between old and new faces-including new faces that contain old parts-through a unidimensional familiarity signal. In the absence of holistic processing, face recognition is based on relational representations that are relatively difficult to encode, but which allow flexible recognition decisions based on match-mismatch detection to be made. Unlike unitized representations, relational representations can support judgments that newly encountered faces match previously experienced faces in some respects (e.g., some of their features) and yet not in others (e.g., other features, global configuration). Four experiments clarified the relationship of holistic processing to the formation of unitized and relational representations of faces. By manipulating the extent of holistic processing while controlling for the overall level of recognition performance, we demonstrate qualitative effects of holistic processing on how recognition decisions are made with faces. (PsycINFO Database Record (c) 2019 APA, all rights reserved).