A first approximation of a model for the automatic synthesis of Italian intonation is proposed. In line with Pierrehumbert's theory of intonational description [1] fundamental frequency contours of Italian are modelled as sequences of abstract tonal elements aligned with the text. Different levels of prosodic phrasing (accent units, intermediate and intonational phrases) are taken into account in the intonational specification of a text string. The theoretical assumptions as well as the two stages ("text-to-phonology" and "phonology-to-phonetics") of the model's implementation in the text-to-speech system are described.