Background.The incidence of schizophrenia is commonly estimated by screening for psychosis among subjects presenting to psychiatric services. This approach (using a first-contact sampling frame) cannot account for cases that did not meet criteria for schizophrenia at first contact. We compared the usual approach directly with a register-based approach (using a longitudinal sampling frame) that also includes subjects initially diagnosed with other non-schizophrenic disorders.Method.We compared data from the Longitudinal Psychiatric Register (LPR) of The Hague over 1980–2009 with data previously collected in a first-contact study, and applied both methods to calculate the incidence of schizophrenia for subjects aged 20–54 years in the same catchment area and over the same period (October 2000 to September 2005). We reconstructed treatment pathways and diagnostic histories up to the end of 2009 and performed sensitivity analyses.Results.The LPR identified 843 first onsets of schizophrenia, corresponding to a treated incidence rate (IR) of 69 per 100 000 person-years [95% confidence interval (CI) 64–74]. The first-contact study identified 254 first onsets, corresponding to a treated IR of 21 per 100 000 person-years (95% CI 18–23). Two-thirds of the difference was accounted for by subjects treated for other disorders before the onset of psychosis, and by patients in older age groups.Conclusions.The incidence of schizophrenia was three times higher in a longitudinal register study than in a high-quality first-contact study conducted in the same population. Risk estimates based only on first-contact studies may have been affected by selection bias.