This paper describes the AT&T Bell Laboratories Mandarin Chinese concatenative text-to-speech (TTS) system, which is being constructed using the NewTTS architecture for multi-lingual TTS [12]. This Mandarin TTS system is 'corpus-based' in that important components of the system are developed on the basis of the analysis of large text corpora. We concentrate here primarily on two modules, namely the word segmentation module and the segmental duration module. We also briefly describe the intonation module and the concatenative unit inventory.