Delusional beliefs and their behavioral consequences are predominant symptoms in patients with psychosis and play an important role in the treatment. Delusional beliefs are a multidimensional concept which can be divided into three components: distress, preoccupation and conviction of delusions. These can be measured using Peters delusions inventory (PDI-21). We question, whether changes in delusional beliefs over time during treatment measured with the PDI-21 can predict changes in belief flexibility measured with the Maudsley assessment of delusions schedule (MADS). We used a group of patients from a randomized controlled trial for a cognitive intervention for psychosis or psychotic symptoms. Aside standard treatment for psychosis, half of the patients took part in a group treatment "Michael's game". Patients were assessed at baseline (T1), at 3 months (T2), and at 9 months (T3). We measured delusional beliefs using PDI-21 and belief flexibility with the MADS. One hundred seventy-two patients were included in the analysis. We measured a main effect of PDI-21scores on belief flexibility measured with MADS. PDI-21 Conviction scores predicted outcomes for all measured MADS items. Increasing PDI Distress and Preoccupation scores were predictors for being more likely to dismiss beliefs and change conviction. Time itself was a predictor for changing conviction and being able to plan a behavioral experiment. Overall the changes in PDI scores predicted outcomes for belief flexibility measured with MADS items. The PDI-21 could be a simple and effective way to measure progress in treatment on delusional beliefs.