In　case-based　reasoning　(CBR),　the　weights　of　feature　attributes　directly　affect　the　quality　of　problem　solving.　This　paper　proposes　a　membrane　computing　(MC)-based　approach　to　optimize　the　attribute　weights.　A　cell-like　membrane　structure　with　three　layers　is　designed.　The　initial　weight　objects　are　then　obtained　by　a　global　search　using　selection,　crossover,　mutation,　and　a　two-way　communication　rule.　Subsequently,　the　best　weight　object　is　obtained　using　a　simulated　annealing　(SA)　algorithm　between　the　membranes.　In　this　manner,　a　group　of　weight　objects　is　received　for　CBR　problem　solving.　The　experiment　results　show　that　the　classification　accuracy　of　this　method　is　higher　compared　with　the　entropy　method,　genetic　algorithms,　SA,　and　the　neural　networks　method.　The　application　of　MC　can　obtain　more　reasonable　attribute　weights,　which　can　effectively　improve　the　quality　of　problem　solving　for　a　CBR　system.　(C)　2014　Elsevier　Inc.　All　rights　reserved.