Hospital in-patient statistics are an important outcome measurement in the assessment of the morbidity associated with diabetes mellitus. A prospective study of 157 consecutive admissions over a 28-day period compared diagnoses obtained from the clinical records with the ICD9 coding of the same admissions recorded at the Information and Statistics Division of the Scottish Health Service. Sixty-one percent of all discharge summaries omitted the diagnosis of diabetes. Even when admission was principally related to diabetes complications, 47% of medical and 88% of surgical discharge summaries omitted diabetes as a diagnostic category. ICD9 coding underestimated the percentage of admissions accounted for by diabetic patients by 100% (2.8 vs 5.6%) and as a result underestimated bed occupancy by over 200% (4.3 vs 13.7%), and is thus failing to fulfil its potential as a demographic and epidemiological record of resource use by disease classification.