A traditional-ordered Tibetan dictionary based on present Tibetan coded character sets (ISO/IEC 10646-1:1993 & GB16959-1997) is of hashing structure, and can make no effective index work because of lacking of ordered internal coded character within computers. This paper establishes a transformational relationship between Tibetan letters and numerical codes with the supplement of analyzing the constructional rules of Tibetan words. According to the statistical analysis of syllabic distribution in a large Tibetan dictionary, we design a multi-level index optimizing project for dictionary data retrieval. The core content includes the idea of layer upon layer processing to the letters of basic consonants and vowels and the matching method based on code prefixes of words. At last we propose a concept of 揵ucket?to process the homographs encountered in data retrieval.
