From Features to Identity: Evidence for a Hierarchical Model of Unfamiliar Face Matching

How does nominal identity emerge from featural information in a face-matching task with unfamiliar faces? The present study extends the classic same-different paradigm to the domain of unfamiliar face matching and proposes a hierarchical serial self-terminating model. The model assumes that similarity between images of faces can be assessed at three representational levels-featural, physical, and nominal-and that observers compare faces sequentially from lower to higher levels, guided by instructional set (physical vs. nominal) and terminating once a discriminating feature is found. Two experiments tested these predictions with upright (Experiment 1) and inverted (Experiment 2) faces. Upright faces showed key diagnostic effects consistent with the model: the absence of a fast-same effect, the presence of the name-physical disparity effect, and costs of nominal relative to physical instructions. Signal-detection analyses confirmed that these reflected changes in sensitivity rather than bias. In contrast, inversion disrupted this pattern, attenuating or reversing effects. The results provide strong but qualified support for a hierarchical account of face matching and highlight the boundary conditions of this architecture.