Objective: To estimate the effect of insight on time to relapse and readmission and on social function and symptoms after following up a cohort of first-episodes of nonaffective psychosis for 18 months. Method: Patients with first episodes of DSM-IV schizophreniform disorder, schizophrenia, schizoaffective disorder, delusional disorder, and psychosis not otherwise specified (excluding primary substance-induced or organic psychoses), aged 16 to 65 years, were recruited over the 26 months from July 1996 to September 1998 from consecutive admissions to day-patient and inpatient units in England with a catchment area population of 2.3 million. They were interviewed with the Positive and Negative Syndrome Scale, Birchwood Insight Scale, and Social Functioning Scale at baseline and 18 months. Results: The hazard ratio for relapse, per unit increase in the insight score, was estimated in a Cox proportional hazards model to be 0.943 (95% CI = 0.892 to 0.996; p = .035). Those with the best insight scores had an estimated rate of relapse that was 39% of that of those with the worst scores (95% CI = 16% to 93%). Readmission was highly correlated with relapse, so poor insight also predicted readmission (hazard ratio 0.934; 95% CI = 0.876 to 0.996; p = .036). However, insight did not independently predict symptoms or social function after adjustment for other predictors of outcome. Conclusion: Insight predicted both relapse and readmission. The details of the beliefs and assumptions determining outcome remain unclear, but intervening to alter them appears to be justified.