Similarity calculation is an important issue in the fields such as the question answering system, information retrieval and so on. This paper designs a model which fuses sentence surface information and the syntax structure for similarity calculation. In the calculation of the sentence similarity, firstly, it considers sentence surface information with two factors of the number of the same keywords and keywords order. Then it considers the similarity of syntactic structure. Finally, the two are combined to calculate the similarity of the sentence for answer extraction. Based on the above algorithms, a facing web intelligent QA system/retrieval system is developed in Java EE platform, Experiments show that the algorithm can effectively improve the performance of the system.