AbstractThis study deals with a set of coding directives that were conceived for trained coding clerks and rely upon knowledge of their cultural background. These directives were formalized and adapted for computer use and in this form must rely upon a background of explicit medical knowledge. Medical data on death certificates are an invaluable source of information regarding prevention of major causes of death. These causes are coded and tabulated worldwide by means of the International Classification of Diseases (ICD). The ICD manual issues directives to achieve uniformity of coding throughout the world. The coder is required to trace back the flow of events which caused death and to single out the most significant concept from the statistical point of view. After emphasizing the problems encountered in the formalization, the methodological contribution of this work to the identification of a modular architecture for a system which represents and “reshapes” knowledge from medical documents is presented. Therefore we focus on the features of the two kinds of knowledge that must be supplied to a knowledge-based system, in order to enable it to perform semantic conversions on given medical data, namely: i) generic guidelines; ii) detailed medical knowledge.