Social anxiety (SA) is associated with a tendency to interpret social information in a more threatening manner. Most of the research in SA has focused on unimodal exploration (mostly based on facial expressions), thus neglecting the ubiquity of cross-modality. To fill this gap, the present study sought to explore whether SA influences the interpretation of facial and vocal expressions presented separately or jointly.Twenty-five high socially anxious (HSA) and 29 low socially anxious (LSA) participants completed a forced two-choice emotion identification task consisting of angry and neutral expressions conveyed by faces, voices or combined faces and voices. Participants had to identify the emotion (angry or neutral) of the presented cues as quickly and precisely as possible.Our results showed that, compared to LSA, HSA individuals show a higher propensity to misattribute anger to neutral expressions independent of cue modality and despite preserved decoding accuracy. We also found a cross-modal facilitation effect at the level of accuracy (i.e., higher accuracy in the bimodal condition compared to unimodal ones). However, such effect was not moderated by SA.Although the HSA group showed clinical cut-off scores at the Liebowitz Social Anxiety Scale, one limitation is that we did not administer diagnostic interviews. Upcoming studies may want to test whether these results can be generalized to a clinical population.These findings highlight the usefulness of a cross-modal perspective to probe the specificity of biases in SA.