According　to　color　and　geometric　attributes　of　traffic　signs　in　China,　an　efficient　traffic　sign　recognition　system　applying　to　natural　scenes　is　proposed　in　this　paper.　In　this　system,　an　improved　image　segmentation　algorithm　based　on　RGB　color　space　is　introduced　to　segment　and　extract　possible　regions　of　traffic　signs　in　natural　scene.　Moreover,　a　two-level　neural　network　is　used　to　classify　and　recognize　traffic　signs,　respectively.　Outline　features　and　moment　invariants　are　used　as　inputs　of　classification　neural　network　and　recognition　neural　network,　respectively.　The　experimental　results　demonstrate　that　the　system　is　capable　of　achieving　a　good　recognition　for　traffic　signs　in　natural　scene;　furthermore,　it　has　high　robustness　and　broad　applicability.