The aims of the study were: (i) to identify trends in the underlying cause-of-death statistics that are due to changes in the coders' selection and coding of causes, and (ii) to identify changes in the coders' documented registration principles that can explain the observed trends in the statistics. 31 Basic Tabulation List categories from the Swedish national cause-of-death register for 1970-1988 were studied. The coders' tendency to register a condition as the underlying cause of death (the underlying cause ratio) was estimated by dividing the occurrence of the condition as underlying cause (the underlying cause rate) with the total registration of the condition (the multiple cause rate). When the development of the underlying cause rate series followed more closely the underlying cause ratio series than the multiple cause rate series, and a corresponding change in the registration rules could be found, the underlying cause rate trend was concluded to be due to changes in the coders' tendency to register the condition. For thirteen categories (fourteen trends), the trends could be explained by changes in the coders' interpretation practice: five upward, four insignificant, and five downward trends. In addition, for three categories the trends could be explained by new explicit ICD-9 rules.