Though exposure-based treatments for social anxiety disorder (SAD) are efficacious, not everyone benefits, and we still do not fully understand how and under what conditions this treatment works. Fruitful approaches to better understanding and improving exposure therapy 1) identify predictors of treatment outcome based on theory and 2) manipulate proposed predictors to improve outcomes. Theoretical models of exposure therapy implicate reduction in threat-related avoidance (or, increases in threat-related attention and approach) as a mechanism of fear reduction. The studies in this dissertation follow this theory-driven approach by testing the hypothesized relation between changes in implicit measures of threat-related attention and treatment outcome (Studies 1 and 2) and testing the efficacy of a strategy aimed to manipulate threat approach to improve exposure outcomes (Study 3). Specifically, in Study 1 we examined how social threat-related attentional biases change during exposure therapy for SAD, and attempted to relate these changes in attention bias with symptom change. Additionally, we tested a computational method used to yield attention bias scores that may be more reliable for repeated measurement. Our results indicated that while social threat-related attention bias did decrease over the course of exposure therapy, these changes were not predictive of symptom change. Study 2 examines the language used during speech exposures (i.e., linguistic style) as a potential predictor of exposure outcome, relating these variables to SAD symptom severity a week after a single exposure session. We found that increased use of first-person plural pronouns during the public speaking exposure was predictive of superior exposure outcomes, in line with an hypothesis that first-person plural pronoun is indicative of threat-related attentional focus during exposure therapy. Finally, Study 3 tested whether a behavioral manipulation thought to temporarily increase endogenous testosterone (a hormone associated with social threat approach) would augment exposure therapy. We found no evidence to suggest that testosterone levels can be manipulated via our behavioral strategy and were thus unable to test whether increasing testosterone facilitates exposure therapy outcomes. We discuss some limitations to the research, as well as recommendations for future directions.