Many question answering systems are now focusing on the research of question-based answer mining from the Web information or snippets returned by the search engine. However, their main goal is only to get the related answers instead of the correct answers because answers are inconsistent. An answer fusion model and method is proposed, which makes semantic analysis of the various answers to the same question returned by search engines or question answering systems and return the fused answer based on fusion rules and data quality. Experimental test in our QA system shows that the method is effective.