OBJECTIVE: This study extended a classic self-referential learning paradigm by investigating the effects of intranasally-administered oxytocin in high and low socially anxious participants during social learning, as a function of social anxiety levels and sex. METHODS: In a randomized double-blinded design, 160 participants were either given intranasal oxytocin (24 I.U.) or placebo. Subsequently, while lying in an MR scanner, participants were shown neutral faces that were paired with positively, neutrally, or negatively valenced self-referential sentences, during which we measured self-reported arousal and sympathy of the facial stimuli, pupil dilation, and changes in the brain-oxygen-level dependent signal. Four-factor mixed analyses of variance with the between-subjects factors group (high socially anxious vs. low socially anxious), substance (oxytocin vs. placebo), and sex (male vs. female) and the within-subjects factor sentence valence (positive vs. neutral vs. negative) were conducted for each measure, respectively. RESULTS: Administration of intranasal oxytocin yielded an increase in sympathy ratings in high socially anxious compared to low socially anxious individuals and decreased arousal ratings for positively-conditioned faces in low socially anxious participants. As an objective physiological measure of arousal, pupil dilation mirrored the behavioral results. Oxytocin effects on neural activation in the insula interacted with anxiety levels and sex: low socially anxious individuals yielded lower activation under oxytocin than placebo; the converse was observed in high socially anxious individuals. This interaction also differed between sexes, as men yielded higher activation levels than women. These findings were more prominent for positively- and negatively-conditioned faces. Within the amygdala, high socially anxious men yielded higher activation than high socially anxious women in the left hemisphere, and low socially anxious men yielded higher activation than low socially anxious women from positively- and negatively-conditioned faces, though no influence of oxytocin was detected. CONCLUSION: These results suggest oxytocin-induced behavioral, physiological, and neural changes as a function of social learning in socially low and high anxious individuals. These findings challenge the amygdalocentric view of the role of emotions in social learning, instead contributing to the growing body of findings implicating the insula therein, revealing an interaction between oxytocin, sex, and emotional valence. Such discoveries raise an interesting set of questions regarding the computational goals of regions such as the insula in emotional learning and how neural activity can play a diagnostic or prognostic role in social anxiety, potentially leading to new treatment opportunities that may combine oxytocin and neurofeedback differentially for men and women.