The purpose of this study was to test cognitive and interpersonal models for improving social phobia. Eighty patients with social phobia were randomized to 10-week residential cognitive (RCT) or residential interpersonal psychotherapy (RIPT). They completed process measures every Thursday and a sub-outcome measure every Monday. The ratings were analyzed with mixed models. Weekly changes in the process variables derived from the cognitive model (self-focus, estimated probability and estimated cost of negative social events, safety behaviors) predicted subsequent weekly changes in social anxiety. Changes in the interpersonal variable perceived acceptance by others also predicted subsequent changes in social anxiety. On the other hand, changes in social anxiety predicted changes in the four cognitive variables. There were no interactive effects of process with treatment. The cognitive variables decreased during treatment to a similar degree in both treatments. The results indicate that, to reduce social anxiety, therapy should target self-focus, estimated probability and cost of feared social events, safety behaviors, and perceived acceptance by others. The process of improvement may involve positive cycles in that a reduction of social anxiety, in turn, appeared to impact self-focus, probability, cost, and safety behaviors.