The main aims were to determine how much diabetes is under-recorded on death certificates, the influence of demographic variables and to what extent New Zealand Health Information Service (NZHIS) coding is able to compensate for deficiencies. A retrospective review of 600 death certificates (during 1999) from the Mortality Review Database in Christchurch Hospitals was carried out. Clinical data, laboratory results, coroner's reports, hospital clinical coding and NZHIS coding were collated and analysed. One hundred and four cases (17%) had clinically documented diabetes, of which only 47 (45%) had diabetes recorded on either the death certificate or the coroner's report. Diabetes was recognised in NZHIS coding in 45 (43%) of cases and in 94 (90%) cases in hospital clinical coding. Diabetes was more likely to be recorded on the death certificate in males and in subjects treated with insulin (P<0.05). Of the 496 subjects without documented diabetes, 159 (32%) had a highest random plasma glucose >11.1 mmol/l. Documented diabetes is thus under-reported on more than 50% of death certificates and this is not compensated by NZHIS coding. In addition, there is likely to be a significant number of cases with undiagnosed diabetes. This may adversely affect the reliability of mortality statistics.