BACKGROUND: Despite growing evidence of the effectiveness of cognitive- behavioural therapy (CBT) for psychosis, typically only about 50% of patients show a positive response to treatment. This paper reports the first comprehensive investigation of factors which predict treatment outcome. METHOD: In a randomised controlled trial of CBT for medication- resistant psychosis (see Part I) measures were taken at baseline of demographic, clinical and cognitive variables. Changes over time were assessed on the Brief Psychiatric Rating Scale and the relationship between potential predictor variables and outcome was investigated using analysis of variance and covariance. RESULTS: A number of baseline variables were identified as predictors of good outcome in the CBT group. Key predictors were a response indicating cognitive flexibility concerning delusions (P = 0.005) and the number of recent admissions (P = 0.002). Outcome was less predictable in the control group and was not predicted by any cognitive variable. CONCLUSIONS: Good outcome is strongly predicted in patients with persistent delusions by a cognitive measure, while this was not the case in controls. Thus we argue that positive outcome in CBT is due in part to specific effects on delusional thinking.