In　view　of　the　problem　that　Hopfield　neural　network　is　easy　to　fall　into　local　minimum,　this　paper　proposes　a　modified　Hopfield　neural　network　to　overcome　this　problem.　By　adjusting　the　connection　weights　which　is　governed　by　the　objective　function,　the　energy　function　decreases　until　to　the　global　minimum.　The　modified　Hopfield　neural　network　is　used　for　solving　TSP　problem,　and　the　experimental　results　show　that　the　network　will　escape　the　local　minimum.　The　validity　and　feasibility　of　the　modified　Hopfield　neural　network　is　proved.