Background <p>Individuals with social anxiety underestimate their social performance compared to observers (i.e., self-observer discrepancy). Cognitive models suggest that heightened attention to, and exaggerated perception of, physiological symptoms of anxiety results in the construction of (inaccurate) mental representations of their external appearance. This study examined the relative contribution of subjective physiological arousal, self-focused attention and objective physiological arousal in predicting self-observer discrepancies during a speech performance task.</p> Method <p>Participants with high (<i>N</i> = 39) and low (<i>N</i> = 36) levels of speech anxiety completed a 10-minute speech task, with continuous measurement of objective physiological arousal (heart rate and skin conductance), self-reported subjective physiological arousal and focus of attention. Speech performance quality was rated by participants and observers to assess self-observer discrepancy.</p> Results <p>There was higher self-observer discrepancy, self-focused attention and subjective physiological arousal in the high vs. low anxiety group, but no significant difference in objective physiological arousal. Subjective physiological arousal and self-focused attention predicted self-observer discrepancy, but objective physiological arousal did not.</p> Conclusions <p>Results are consistent with cognitive models of social anxiety suggesting that cognitive and attentional biases, rather than differences in objective physiological experience, are key in understanding self-observer discrepancies in social anxiety.</p>