Attentional biases in the processing of threatening facial expressions in social anxiety are well documented. It is generally assumed that these attentional biases originate in an evaluative bias: socially threatening information would be evaluated more negatively by socially anxious individuals. However, three studies have failed to evidence a negative evaluative bias in the processing of emotional facial expression (EFE) in socially anxious individuals. These studies however suffer fromseveral methodological limitations that the present study has attempted to overcome. Twenty-one out-patients diagnosed with generalized social phobia have been compared to 20 out-patients diagnosed with another anxiety disorder and with 39 normal controls matched for gender, age and level of education. They had to decode on seven emotion intensity scales a set of 40 EFE whose intensity and emotional nature were manipulated. Although sufficient statistical power was ensured, no differences among groups could be found in terms of decoding accuracy, attributed emotion intensity, or reported difficulty of the task. Based on these findings as well as on other evidences, we propose that, if they exist, evaluative biases in social anxiety should be implicit and automatic and that they might be determined by the relevance of the stimulus to the person's concern rather than by the stimulus valence. The implications of these findings for the interpersonal processes involved in social phobia are discussed.