Positive self-beliefs are important for well-being, and are influenced by how others evaluate us during social interactions. Mechanistic accounts of self-beliefs have mostly relied on associative learning models. These account for choice behaviour but not for the explicit beliefs that trouble socially anxious patients. Neither do they speak to self-schemas, which underpin vulnerability according to psychological research. Here, we compared belief-based and associative computational models of social-evaluation, in individuals that varied in fear of negative evaluation (FNE). Using a novel analytic approach, ‘clinically informed model-fitting’, we replicated the finding that high-FNE participants learn faster from negative feedback about themselves. Crucially, this could be explained through reduced activation of positive self-schemas. The overall population could be characterized equally well by belief-based or associative models, but many individuals used either the one or the other perspective. Our findings have therapeutic importance, as belief activation may be used to specifically modulate learning