We examined the predictive value of the discharge diagnoses of stroke and transient ischemic attack (TIA) in The National Registry of Patients (NRP) for participants in the Danish cohort study "Diet, Cancer, and Health." We retrieved all probable incident registered cases of stroke and TIA, i.e., ICD-10: I60-69.8, or G45 (n = 581) within the cohort from the NRP. Medical records and hospital discharge summaries were retrieved and reviewed using a standardized form. Overall, 299 of 377 cases (79.3%, 95% CI: 74.9-83.3%) of stroke recorded were confirmed. Subarachnoidal hemorrhage and intracerebral hemorrhage were confirmed in 14 of 29 cases (48.3%, 95% CI: 29.4-67.5%), and 23 of 35 cases (65.7%, 95% CI: 47.8-80.9%), respectively. By contrast, ischemic stroke and unspecified stroke were confirmed in 99 of 113 cases (87.6%, 95% CI: 80.1-93.1%) and 152 of 200 cases (76.0%, 95% CI: 69.5-81.7%), respectively. Among 134 patients with a TIA discharge diagnosis, 60.4% (95% CI: 51.6-68.8%) were confirmed. Discharge diagnoses from emergency rooms had lower overall predictive value (48.8%, 95% CI: 39.9-57.8%) than discharge diagnoses from departments of internal medicine (68.8%, 95% CI: 61.3-75.5%) and departments of neurology or neurosurgery (77.9%, 95% CI: 72.3-82.7%). We conclude that stroke and TIA diagnoses in NRP should be used with caution in epidemiological research because the low predictive value for some diagnostic subgroups may lead to serious misclassification and biased results.