Participants heard 10 voices and were given a 2-alternative recognitio
n-memory test. Accuracy was higher when testing was intentional rather
than incidental and when faces accompanying the original voices were
re-presented during the test. Some support was found for an attention-
allocation model, which predicted poorest performance with incidental
testing and faces not reinstated. Reported confidence was higher for c
orrect than for incorrect decisions, but more confident participants w
ere not more accurate.