Two-stage　fuzzy　controller,　which　is　easy　to　acquire　fuzzy　rules　and　can　greatly　decrease　the　number　of　fuzzy　rules,　is　very　suitable　to　complex　urban　intersection　control.　But　it　still　presents　a　difficulty　for　deciding　all　the　membership　functions　correctly　only　by　human　experience.　There　for,　to　solve　this　problem,　a　technique　using　genetic　algorithm　to　modify　its　fuzzy　membership　functions　is　proposed.　For　a　single　urban　intersection　with　four-phase,　much　simulation　is　processed.　Simulation　result　shows　that　this　method　can　effectively　decrease　the　average　delaying　time　in　the　intersection　and　is　superior　to　the　conventional　method.