1.136 death certificates representing all 1985 Bremen cardiovascular deaths and a 50%-sample of non-cardiovascular deaths in the age group 25-69 years were analyzed for reliability of nosologists' coding according to ICD-coding rules (9th revision). The 1.136 photocopied death certificates were used to assess intra-observer-variation in Bremen and to determine inter-observer-variation among 7 nosologists from 6 different State Statistical Offices and the Federal Statistical Office. Intra-observer-agreement in Bremen was found to be similar to the results presented in a comparable US-study: Bremen: 92.1%; Curb et al. 1983: 94.8%-96.1%; 3-digit-ICD-Code. Inter-observer-agreement was found to be much lower in Germany than in two US-studies: 3 coders agreeing on 3-digit-ICD-Code: Bremen: 67.7% (average, 3 coders out of 7); Curb et al.: 90.2% (3 coders); 3 coders agreeing on 4-digit-ICD-Code: Bremen: 61.5%; NCHS 1980: 90.3%. Agreement-rates were also much lower in Germany than in the USA (Curb et al.) when particular disease groups were analysed: Ischaemic heart disease (ICD 410-414): Bremen: 82.7% (average); USA: 97.2%; cerebrovascular disease (ICD 430-438): Bremen 65.6% (average); USA: 93.2%; neoplasms (ICD 140-239): Bremen: 94.0% (average); USA: 97.8%. We conclude that training, individual characteristics of nosologists, and other factors may cause important artifacts when comparing German mortality statistics on a regional level or during different time intervals.