AIM: Duration of untreated psychosis (DUP) is considered as a key prognostic variable in psychosis. Yet, it is unclear whether a longer DUP causes worse outcomes or whether reported associations have alternative explanations. METHODS: Data from 2 cohorts of patients with first episode psychosis were used (n = 2134). Measures of DUP were assessed at baseline and outcomes at 12 months. Regression models were used to investigate the associations between DUP and outcomes. We also investigated whether any associations were replicated using instrumental variables (IV) analysis to reduce the effect of residual confounding and measurement bias. RESULTS: There were associations between DUP per 1-year increase and positive psychotic symptoms (7.0% in symptom score increase 95% confidence interval (CI) 4.0%, 10.0%, P < .001), worse recovery (risk difference [RD] 0.78, 95%, CI 0.68, 0.83, P < .001) and worse global functioning (0.62 decrease in functioning score 95% CI -1.19, -0.04, P = .035). There was no evidence of an association with negative psychotic symptoms (1.0%, 95%, CI -2.0%, 5.0%, P = .455). The IV analysis showed weaker evidence of associations in the same direction between DUP per 1-year increase and positive psychotic symptoms, recovery and global functioning. However, there was evidence of an inverse association with negative psychotic symptoms (decrease of 15.0% in symptom score 95% CI -26.0%, -3.0%, P = .016). CONCLUSIONS: We have confirmed previous findings of a positive association between positive psychotic symptoms, global functioning and recovery and DUP using regression analysis. IV analysis shows some support for these findings. Future investigation using IV analysis should be repeated in large data sets.