The agreement between self-reported diseases in a questionnaire survey and data from medical records was assessed in a representative sample of Finnish men and women (n = 596) aged 45-73 years. The accumulated medical record information was abstracted from the records in the health centers and the central hospital in the study region. The agreement between the two information sources was substantial (kappa 0.73-0.80) for cardiovascular diseases as a group, hypertension, angina pectoris, myocardial infarction, and diabetes. The lowest agreement (kappa < 0.55) was found for lower back disorder, hip and knee arthrosis, and claudication. These results showed that the agreement between questionnaire data and medical records was good for well-known chronic diseases that have clear diagnostic criteria and are easily communicated to the patient. Conversely, the agreement was poor for diseases with nonestablished diagnostic criteria and a fluctuating course.