A similarity-based filtering algorithm is proposed for multi-level matching. The algorithm consider all elements like service name, description information, input and output, reputation and effects respectively, cosine theorem and semantic distance are given in this paper to calculate their similarity. The aspects of the algorithm are as follows. It achieves a more comprehensive metrical scheme on the similarity between publishing service and request service from multiple stages and multiple perspectives in web service matching process. It judges whether the service preconditions match the effects based on description logic. Finally, it introduces reputation metrical scheme .In addition, the qualitative and quantitative analysis of the algorithm are given in this paper. The analysis results show that the Multi-level Matching Filtering Algorithm can obviously improve the recall ratio and precision ratio of Web service discovery.