Corpus-based speech synthesis is the most widely-used speech synthesis technology in home and abroad. In this type of synthesis method, unit selection is crucial to the speech synthesis. This paper designs a system model of Tibetan speech synthesis by analyzing the attributive characteristics of Tibetan text, and presents a mixed units mode with Tibetan components, combinational components, characters, words and sentences. The method effectively preserves the integrity of larger units and the flexibility and robustness of small units. At the same time, it provides the unit selection strategies and algorithms of Tibetan speech synthesis. Experimental data indicates that the strategies and algorithms are effective and the coverage of units reaches expected target in both the open corpus and closed corpus.