Most of the inverted index structures do not refer to the organization of the word table, and there are lots of word collisions in the conventional Hash algorithms. This paper proposes a twolevel index structure, which uses simply a coding method to map words beginning with the same word to the same position of the first level index, and uses a shareprefix tree as the second level index to find the address of the index files rapidly, and reduces the storage space of the index files. The experimental results show that, the compressing ratio of the size of index files to that of the source files reaches 0.59. Compared with the sequence index and the Hash index, we acquire a better spaceandtime efficiency.