Self-help and psychoeducation have been identified as effective methods for delivering treatment, yet not everyone benefits from these brief interventions. Therefore it is clinically and economically useful to identify who is likely to require more intensive assistance. This paper develops a prognostic scale which predicts who will recover from panic attacks and who will require more assistance. Method: Random regression models were used to evaluate the relationship between predictive variables, baseline severity, and the rate of improvement in 117 people with DSMIV panic attacks who participated in a trial of a psycho-educational booklet, a self-help workbook, and brief group CBT over a 9-month period. ROC analysis was used to choose cut-off points on a scale made up of significant predictors. Results: Panic disorder and agoraphobia symptom measures were predicted by baseline social anxiety, and general mental health. There was no significant effect on the outcome for baseline depression or anxiety sensitivity. While general mental health (SF12 Mental Component scores) was predicted by the age at first panic attack, neuroticism, panic disorder and/or agoraphobia symptoms and a positive screen for alcohol use disorders. A prognostic scale based on simple additive scoring was equivalent to standard scores and significantly better than chance at predicting who would recover and who required face-to-face therapy. Conclusions: The prognostic scale may be used to guide the choice of psychoeducation, self-help or face-to-face therapy as the first step in stepped care.