A small body of research suggests that socially anxious individuals show biases in interpreting the facial expressions of others. The current study included a clinically anxious sample in a speeded emotional card-sorting task in two conditions (baseline and threat) to investigate several hypothesized biases in interpretation. Following the threat manipulation, participants with generalized social anxiety disorders (GSADs) sorted angry cards with greater accuracy, but also evidenced a greater rate of neutral cards misclassified as angry, as compared to nonanxious controls. The controls showed the opposite pattern, sorting neutral cards with greater accuracy but also misclassifying a greater proportion of angry cards as neutral, as compared to GSADs. These effects were accounted for primarily by low-intensity angry cards. Results are consistent with previous studies showing a negative interpretive bias, and can be applied to the improvement of clinical interventions.