Fig. 1. Design and behavior.

a Experimental conditions: participants learned pairs of faces, either a famous and a novel face in the prior-knowledge condition (PK), or two novel faces in the no-prior-knowledge condition (n-PK). b While in the scanner, participants viewed the pairs 12 times in 12 cycles; each cycle included all pairs in a random order. Before and after associative learning, participants in the scanner also viewed each face presented alone, in random order. This allowed us to capture the multivoxel activity pattern of each face, for pattern similarity analysis (see Methods). c After the post-learning scan, participants performed an associative memory test, in which they indicated which of the three bottom faces appeared with the face on the top and rated their confidence (sure/possibly/maybe). d Behavioral results of the associative memory test. High-confidence hits include “sure” and “possibly” responses. **P = 0.01, the results of a paired-sample two-tailed t test between PK and n-PK pair type. Data are presented as mean values, error bars reflect +/− SEM. To protect copyrights, all faces in the figures are of novel faces, and we obtained permission to use these photos. Participants in the study saw famous and novel faces, as detailed in Methods.