High accuracy automatic segmentation of Mandarin TTS (text to speech) corpus is vital for obtaining high quality syllable’s boundary to corpusbased speech synthesis. Among the existing methods, most studies on automatic segmentation are based upon single model, ignoring the diverse time marks gained by different models in specific Mandarin boundary environment. In this paper, three hidden Markov models (HMM) models initial-final monophonebased HMM, semi-syllable monophone-based HMM and initial-final triphonebased HMM are adopted respectively. Making use of decision tree algorithm C4.5, a classification approach is proposed to combine the advantages of the three models together by means of selecting the most appropriate one for each boundary unit. The experimental results show that the proposed method can achieve better performance than the single model method, in terms of error rate and time shift of boundaries. Keywords: automatic segmentation, Mandarin TTS, decision tree