A register has been built for planned epidemiological studies of sick-leave, containing all cases exceeding 6 days in a population of 184,000, over a period of 3 years. The diagnoses were coded from medical certificates. To assess the quality of this information this study reviews the medical certificates of 2,364 cases. In 299 cases the corresponding medical records are reviewed and independent diagnoses made. The coding and entering of data into the register is correct in 98% of cases. The independently-made diagnoses match exactly the ones registered in 50% of cases. When grouping the diagnoses into 39 groups, the match on group level is 72%. Ten percentage points of the mismatch are caused by specified overlaps between groups. The remaining 18% mismatch is caused mainly by different interpretations or unspecific labelling of the disease states, not so much by them being obscure in themselves or by doctor covering up unpleasant diagnoses.