Aim: To determine to which extent and how coding errors affect the Belgian mortality statistics. Methods: According to rules and notes of the ICD-9 Classification, an automated programme was developed and applied to the Belgian 1980-1997 mortality data. In addition to this, a recoding exercise has been carried out in collaboration with the French WHO Coding Reference Centre on a systematic sample of 566 certificates. Main findings: The number of errors decreased significantly over time (from about 100 °/°° errors to 10 °/°°), but the pattern of errors changed in time and in magnitude, according to region of residence and age group (higher error rates in the higher age group, lower error rates in Brussels). The high proportion of certificates, mentioning only one cause of death, prevented to study these phenomena according to type of pathology. Moreover, we found that the real error rate is much higher than that established by the automated programme as shown by the recoding exercise, which determined a four times higher error rate. Principal conclusions: Notwithstanding the improved quality of our mortality data, a continuous effort of quality assurance and of the timely availability of the statistics is mandatory. It aims essentially at the accuracy of the coding, in which comparisons with other registers are essential, and at the allocation of the necessary resources. Due to shortcomings of the past, one has to be cautious by the interpretation of our older mortality data and of time trends. ; Doelstelling: Bepalen in welke mate en hoe codeerfouten de Belgische doodsoorzakenstatistiek beïnvloeden. Methode: Een geautomatiseerd programmema, waarin regels en nota’s van de ICD-9 classificatie werden geïntegreerd, werd uitgewerkt en toegepast op de Belgische mortaliteitsgegevens van 1980-1997. Daarenboven werd in samenwerking met het Franse WGO Codeer Referentiecentrum een hercoderingsoefening uitgevoerd op een systematische steekproef van 566 certificaten. Resultaten: Over de jaren heen verminderde het aantal ...