Until quite recently research in rhetoric, literary theory and cognitive semantics was characterised by the dominance given to metaphor. Metonymy was seldom subjected to the same type of scrutiny. Until now it "word schromelijk verwaarloosd", as one Dutch investigator expressed it in 1989. This situation seems to be changing rapidly as metonymy is being discovered as a cornerstone of human cognition and ordinary language use (see the Workshop on "Conceptual Metonymies" at Hamburg University, June 23 to 24, 1996). However, "confusion piles upon obscurity when we consider the treatment given to synecdoche" (Bredin 1984:45). Not only is there no clear definition of synecdoche, there is also neither agreement on the various types of synecdoche nor on its relation to metonymy. Bernard Meyer summarises the situation well when he writes: "La catégorie de la synecdoque apparaît donc comme une classe rhétorique d'extension flottante, une nébuleuse de figures variant autour d'un noyau stable." (Meyer 1993:85). We shall see that even this kernel is not stable either. In this paper I shall briefly outline the history of the treatment of synecdoche from antiquity to he present, that is in traditional rhetoric, historical semantics, Jakobson, general rhetoric and cognitive semantics. I shall then look at some modern attempts at defining synecdoche by comparing Hugh Bredin's approach with that proposed by Ken-ichi Seto. And finally, I shall try to outline the place of synecdoche in a theory of pragmatic inference. (Brigitte Nerlich)