Studies using linguistic stimuli have provided little support for explicit memory biases among individuals with social phobia (SP). However, using facial stimuli rated on their criticalness, Lundh and Ost (1996) found that individuals with SP recognized more critical than accepting faces, whereas non-anxious controls tended to show the opposite pattern. Since the publication of Lundh and Ost's findings, additional studies using a variety of facial stimuli have produced inconsistent findings (J. Anxiety Disord. 14 (2000) 501; Behav. Res. Ther. 39 (2001) 967). Unfortunately, these inconsistencies are difficult to reconcile given great variation in methods and stimuli. Therefore, we designed a study to replicate and extend the work of Lundh and Ost (Behav. Res. Ther. 34 (1996) 787). Similar to Lundh and Ost, individuals with SP identified a significantly higher proportion of old critical faces as old than did non-anxious controls. Further, extending the work of Lundh and Ost, signal detection analyses revealed group differences on response bias according to face type. Specifically, controls showed a response bias towards indicating that accepting faces were previously seen, whereas individuals with SP did not. Finally, signal detection analyses failed to reveal group differences in the accuracy of memory.