This paper presents a model of Italian intonation included in a speech synthesis system developed at CSELT (Torino). The model is based on two theoretical assumptions. The first concerns the structure of the intonational system: following Pierrehumbert (1980) and Liberman & Pierrehumbert (1984), we think that intonation is the result of the interaction of principled choices in four different systems: phrasing, tune, pitch accent placement and pitch range. The second concerns the actual shape of F0 contours: we think is determined by categorical choices made from a small set of discrete phonological elements (Ladd, 1987).