In a face-matching task, subjects compare photos of a target person to different photos of the target or to photos of similar-looking foils. People are highly accurate when matching familiar faces but considerably worse when matching unfamiliar faces. Previous studies suggest that unfamiliar face matching improves when observers study multiple high variability images of the target. These results suggest that identifying an unfamiliar face improves when the studied photos contain greater within-person variability. The present study examined face matching accuracy when two high or low variability study images of a target identity were followed by two high or low variability comparison images of either the same or a foil identity. Within-person variability was manipulated across trials. Of interest was the correspondence in variability of the study and comparison images (e.g., matching: low study variability, low comparison variability; mismatching: low study variability, high comparison variability). When a comparison identity was a "match" (same identity) to the previously shown target pair, lower variability in the test array led to better unfamiliar face matching accuracy than high variability, but only when the learning array was also low in variability. In contrast, when a comparison identity was a "mismatch" (different but similar-looking identity) to the previously shown target pair, face matching accuracy was not significantly affected by variability in study or comparison images. Thus, in contrast to suggestions from previous research, high variability does not necessarily improve unfamiliar face matching.