In this paper we propose a multi-tier phonetic model for intonation synthesis. Four tiers are defined aiming at different components of intonation. The parametric representation for each tier is realized through underlying pitch target. Regression trees are built for each tier upon appropriate feature set. Both numerical evaluation and informal listening test have yielded encouraging results. The synthesized sentences can be accessed at http://mel.speech.northwestern.- edu/sunxj/prosody.htm