Registers have been widely used in schizophrenia studies. For bipolar disorder, the use has been more limited and no studies exist concerning accuracy of bipolar I diagnoses in registers. We have collected an unselected twin sample for a study of bipolar I disorder. We report here the diagnostic procedures and accuracy of bipolar I diagnoses in a nation-wide hospital discharge registry. We evaluated also the accuracy of the medical record-based best-estimate diagnoses, comparing these with the SCID interview diagnoses. From the National Hospital Discharge Register we identified all like-sex twins, born 1940-1957, (N=42), and all like- or opposite-sex twins, born 1961-1969, (N=15), who had at least once a diagnosis of bipolar I disorder. The best-estimate diagnoses based on medical records, and were made by two doctors blind to each other, and according to DSM-IV-criteria. Diag-noses were then confirmed by SCID interviews, with the mi-nimum follow-up time after the first admission being 6 years. The accuracy of bipolar diagnoses in the register was 92% and 87%, accordingly. When comparing the best-estimate diagnoses to SCID interview diagnoses, there was one false positive case in the first sample, 3.4% [95% CI 0.1% to 7.8%], and none in the second sample. Our study shows that a diagnosis of bipolar I disorder in the hospital discharge register, or if based only on medical records, is highly reliable and stable. These results strengthen the usefulness of regis-ters also in bipolar disorder.