In　this　paper,　an　improved　Levenberg-Marquardt　(LM)　algorithm　with　adaptive　learning　rate　is　proposed　to　optimize　the　learning　process　of　RBF　neural　networks.　First,　an　improved　LM　algorithm　is　adopted　using　a　quasi-Hessian　matrix　and　gradient　vector　which　are　computed　directly.　Compared　with　the　conventional　LM　algorithm,　Jacobian　matrix　multiplication　and　storage　are　not　required　in　the　improved　LM　algorithm,　which　can　reduce　computation　cost　and　solve　the　problem　of　memory　limitation.　Second,　the　adaptive　learning　rate　is　integrated　into　the　improved　LM　algorithm　in　order　to　accelerate　the　convergence　speed　of　training　algorithm　and　improve　the　network　performance　of　nonlinear　system　modeling.　Finally,　several　experiments　are　conducted　and　the　results　show　that　the　proposed　method　has　faster　convergence　speed　and　better　prediction　performance.　©　2016　TCCT.