A method for sentence similarity computation by integrating multi-features was proposed. According to the syntax feature, semantic feature and word feature of the sentences, the similarity was measured, respectively. Then, this paper combined the sentence similarity by endowing the above three features with different weights. Comparatively, this kind of estimation of sentence similarity is more accurate than the previous because both the deep and surface information of the sentences were taken into accounted, and the vocabulary of sentences was also extended in the process of calculation.