A number of controlled trials have demonstrated the efficacy of Internet-based cognitive-behaviour therapy for treating social anxiety disorder (SAD). However, little is known about what makes those interventions work. The current trial focuses on patient expectations as one common mechanism of change. The study examines whether patients' expectancy predicts outcome, adherence, and dropout in an unguided Internet-based self-help programme for SAD. Data of 109 participants in a 10-week self-help programme for SAD were analysed. Social anxiety measures were administered prior to the intervention, at week 2, and after the intervention. Expectancy was assessed at week 2. Patient expectations were a significant predictor of change in social anxiety (β = - .35 to - .40, all p < .003). Patient expectations also predicted treatment adherence (β = .27, p = .02). Patients with higher expectations showed more adherence and better outcome. Dropout was not predicted by expectations. The effect of positive expectations on outcome was mediated by early symptom change (from week 0 to week 2). Results suggest that positive outcome expectations have a beneficial effect on outcome in Internet-based self-help for SAD. Furthermore, patient expectations as early process predictors could be used to inform therapeutic decisions such as stepping up patients to guided or face-to-face treatment options.