In　this　paper,　a　tabu　based　neural　network　learning　algorithm　(TBBP)　is　represented　to　improve　the　function　approximation　ability　of　neural　networks　to　nonlinear　functions.　By　using　the　tabu　search　during　the　search　process　in　the　global　area,　the　algorithm　can　escape　from　the　local　optimal　solution　and　get　a　superior　global　optimization　for　the　neural　networks.　The　TBBP　is　tested　in　6　different　nonlinear　functions.　It　is　compared　with　the　standard　BP　algorithm.　The　results　show　that　the　tabu　search　has　improved　the　ability　of　the　approximating　ability　of　the　neural　networks.　©　2006　IEEE.