The　attribute　weights　assignment　in　case-based　reasoning　(CBR)　system　may　determine　the　similarities　between　cases,　and　thus　it　has　a　significant　impact　on　the　correctness　of　reasoning.　To　improve　the　reasoning　performance,　the　water-filling　theory　is　introduced　to　the　attribute　weights　assignment　in　this　paper.　Reasonable　indicators　of　weight　distribution　are　established,　an　associated　Lagrange　function　is　constructed　and　the　weight　optimization　solution　can　be　achieved.　Thereby　a　convergent　water-filling　assignment　(WFA)　algorithm　is　obtained　which　can　be　used　in　the　weighted　K-nearest　neighbor　rule　to　retrieve　similar　cases.　Classification　experiments　for　comparison　between　the　mean　assignment　method,　WFA　method　and　genetic　algorithms　for　the　attribute　weights　using　the　5-fold　cross-validation　method　are　conducted.　The　results　show　that　the　classification　performance　of　CBR　can　be　further　increased　after　the　attribute　weights　are　assigned　by　WFA.　Copyright　©　2014　Acta　Automatica　Sinica.　All　rights　reserved.