To the end of improving the naturalness of the intonation generated by a text-to-speech system for Italian, a syntactico-prosodic pre-processor has been devised, inserting prosodic markers at syntactic boundaries in the input text. The relations between syntax and prosody, to be exploited by the system, are being investigated on a speech data base obtained by digital recording of a carefully designed corpus of sentences read by a professional speaker. First results concern the duration and position of breathing pauses, correlated to sentence syntactic structure, A prototype of the syntactico-prosodic parser has been implemented and tested on written texts.