Complete and accurate registration of cancer is needed to provide reliable data on cancer incidence and to investigate aetiology. Such data can be derived from national cancer registries, but also from large population-based cohort studies. Yet, the concordance and discordance between these two data sources remain unknown. We evaluated completeness and accuracy of cancer registration by studying the concordance between the population-based Rotterdam Study (RS) and the Netherlands Cancer Registry (NCR) between 1989 and 2012 using the independent case ascertainment method. We compared all incident cancers in participants of the RS (aged ≥45 years) to registered cancers in the NCR in the same persons based on the date of diagnosis and the International Classification of Diseases (ICD) code. In total, 2,977 unique incident cancers among 2,685 persons were registered. Two hundred eighty-eight cancers (9.7%) were coded by the RS that were not present in the NCR. These were mostly nonpathology-confirmed lung and haematological cancers. Furthermore, 116 cancers were coded by the NCR, but not by the RS (3.9%), of which 20.7% were breast cancers. Regarding pathology-confirmed cancer diagnoses, completeness was >95% in both registries. Eighty per cent of the cancers registered in both registries were coded with the same date of diagnosis and ICD code. Of the remaining cancers, 344 (14.5%) were misclassified with regard to date of diagnosis and 72 (3.0%) with regard to ICD code. Our findings indicate that multiple sources on cancer are complementary and should be combined to ensure reliable data on cancer incidence. © 2019 The Authors. International Journal of Cancer published by John Wiley & Sons Ltd on behalf of UICC.