An analysis is presented of death certification in Dunedin Hospitals. The study involved a comparison of death certificates with post-mortem findings in 643 deaths which occurred in Dunedin hospitals between 1 October 1971 and 30 September 1973. The cases were categorised according to diseases in the various body systems. The study also involved consideration of the format of the current death certificate and of attitudes towards death certification. Serious defects have been revealed in all areas of the study which are considered to be representative of the national scene. Errors of epidemiological significance were found in 64.7 percent of the certificates, and in 57.5 percent of individual diagnoses. Responsibility for the defects lies with the certification itself and not with the statisticians who diligently compile mortality data from the information supplied to them.