Abstract : This study deals with computer representations of musical knowledge on the basis of two real-scale experiments. The first experiment focusses on knowledge acquisition in ethnography: an expert (the musician), an analyst (the musicologist) and a machine are interacting in a learning situation. Improvisation schemata through which musicians express themselves are identified and formalized with production rules in a formalism derived from generative grammars and pattern languages. A deterministic algorithm is introduced for assessing the membership of arbitrary strings to the langage defined by a given (context-sensitive) grammar. A technique for the inductive inference of regular languages is presented, enabling automatic knowledge acquisition of syntactic and lexical knowledge. The second experiment is part of the design of a computer environment for musical composition. Here the problem is time representation in a discrete structure of “time objects”, more generally the synchronization of parallel processes. A method is outlined for the determination of a structure with incomplete data about the synchronization of its objects. The concept of “sound object” is then formally introduced. An efficient algorithm is proposed for the time-setting of objects in a structure, given the constraints arising from their metric and topological properties.