This paper concerns an investigation of outcome predictors in a clinical trial of psychological therapies for generalized anxiety disorder. A variety of information of potential predictive value was obtained at three stages of patient contact: the initial referral, a screening interview and early sessions of therapy. Three measures of the clinical significance of change over a 12-month follow-up period were used to construct a composite measure which categorized outcome in terms of sustained improvement, relapse and no consistent change. Logistic regression was used to examine the validity of predictors identified in previous research and the relative importance of data obtained from the three different stages. Seventy-one per cent of patients were correctly classified as improved or not from initial data with a significant increase in accuracy with information from the screening interview (77 per cent) and early sessions (82 per cent). Patients who relapsed or not were predicted with considerable accuracy from initial data (90 per cent) and there was no significant increase in predictive power with additional information. The most powerful and robust predictors were: type of treatment received, marital status, marital tension and complexity of clinical presentation in terms of axis 1 co-morbidity. A conceptual framework for prediction is outlined.