To solve the problem that users of web search engines are often forced to sift through the long ordered list of document, a new key-feature clustering (KFC) algorithm was presented to help locate the valuable search results that the users really needed, which was different from VSM. The algorithm firstly extracted some key features from the keywords in the search results. Then the relationships between key features were analyzed and features were clustered. Finally, the documents were clustered based on these clusters of key features. The algorithm was tested and validated by the results of experiments.