Abstract
Face detection and recognition are often characterized as efficient and automatic. Nevertheless, the degree to which multiple faces can be processed in parallel is not known. Visual processing capacity limits can be assessed using a divided attention paradigm, in which observers make independent judgments about two simultaneous stimuli. Capacity limits impair performance in a “dual-task”, which requires the processing of both stimuli, as compared to a “single-task” in which only one stimulus must be processed. Sufficient impairment on a dual-task versus a single-task is interpreted as serial processing. Here, we used a divided attention paradigm in which observers viewed two faces simultaneously to the left and right of central fixation and had to match either the identity of the faces or the color of the faces. We compared performance for the two kinds of judgments (identity versus color) in a dual-task condition (observers judge faces on both sides) and a single-task condition (observers judge only one side and ignore the other). An adaptive method was used to equate difficulty between identity and color judgments for individual observers. For identity matching, the dual- and single-task accuracy differed sufficiently to be best explained by a serial processing model. In contrast, color matching judgments showed a minimal cost of dividing attention, as predicted by an unlimited capacity parallel processing model. We conclude that, despite abundant evidence that single faces are processed efficiently and automatically, capacity limits arise for multi-face identity processing, even though non-identity judgments can be performed for two faces in parallel without a cost of dividing attention. Our results complement other recently published findings that challenge the automaticity of face processing beyond single faces.