Background: The validity of dementia diagnoses in the Danish nationwide hospital registers was evaluated to determine the value of these registers in epidemiological research about dementia. Methods: Two hundred patients were randomly selected from 4,682 patients registered for the first time with a dementia diagnosis in the last 6 months of 2003. The patients' medical journals were reviewed to evaluate if they fulfilled ICD-10 and/or DSM-IV criteria for dementia and specific dementia subtypes. The patients who were still alive in 2006 were invited to an interview. Results: One hundred and ninety-seven journals were available for review and 51 patients were interviewed. A registered diagnosis of dementia was found to be correct in 169 (85.8%) cases. Regarding dementia subtypes, the degree of agreement between the registers and the results of the validating process was low with a kappa of 0.36 (95% CI 0.24-0.48). Conclusion: The validity of dementia syndrome in the Danish hospital registers was high and allows for epidemiological studies about dementia. Alzheimer's disease, although underregistered, also had a good validity once the diagnosis was registered. In general, other ICD-10 dementia subtypes in the registers had a low validity and are less suitable for epidemiological research.