Aiming　at　adopting　what　kind　of　strategy　to　argue　in　Argument-based　negotiation　of　agents,　this　paper　adopts　methods　of　Case-Based　Reasoning　(CBR)　and　reinforcement　learning　together　and　put　forward　a　model　for　negotiation　strategy　selection　based　on　Multi-Agent　System　(MAS)　via　constructing　case　base　about　negotiation　strategy　and　searching　in　the　existing　case　base　according　to　the　similarity　of　the　attributes　that　are　proposed　in　advance,　in　order　to　find　the　most　similar　case,　evaluate　the　strength　of　argument　by　reference　the　reusable　case,　and　select　the　negotiation　strategy.　If　there　is　no　reusable　case,　reinforcement　learning　method　will　be　used　to　evaluate　and　select　the　argument　to　send,　and　update　the　case　base.　Relevant　examples　and　prototype　system　is　realized　for　further　analysis　and　testing.　This　method　not　only　has　the　high　efficiency　of　case-based　reasoning　but　also　takes　on　the　dynamic　accommodation　of　reinforcement　learning,　and　perfects　the　mechanism　of　negotiation　strategy　selection.　©　2010　IEEE.