In recognition memory for unfamiliar faces, performance for target-present items (hits) does not correlate with performance for target-absent items (false positives), a result which runs counter to the more usual mirror effect. In this paper we examine subjects' performance on face matching, and demonstrate no relationship between performance on matching items and performance on nonmatching items. This absence of a mirror effect occurs for multidistractor, l-in-10 matching tasks (Experiment 1) and for simple paired-item tasks (Experiment 2). In Experiment 3 we demonstrate that matching familiar faces produces a strong mirror effect. However, inverting the familiar faces causes the association to disappear once more (Experiment 4). We argue that familiar and unfamiliar faces are represented in qualitatively different ways.