To settle the defects that the repetitive and redundant information returned by meta searchengine leads to increase the burden of result display agent and reduce the precision, a meta search en-gine system model based on mufti-agent is brought forward.The importance degree of membersearch engines to a particular query comprehensively is measured, in the aspects of member Agent'scrawling capability, the relevance between retrieve documents and certain query, and response timeduring query.Then the importance degrees of member search engines are ordered by descend, so asto select several member search engines with higher importance degree to dispatch and merge.It hasproved through the experimental comparison that compared to the traditional meta search engine, theintelligent meta search engine based on reward mechanism improves the retrieval efficiency and queryperformance to a certain extent.