Based　on　Response　Surface　Methodology　(RSM),　the　objective　function　and　constraint　conditions　are　approximately　explicit　to　overcome　difficulty　of　sensitivity　analysis,　the　Sequential　Quadratic　Programming　(SQP)　model　was　formed　to　find　optimal　shape　under　displacement　and　stress　constraints.　An　improved　response　surface　with　exact　value　on　the　design　point　is　constructed,　which　eliminates　the　intrinsic　inaccuracy　of　response　surface.　The　constraint　response　surface,　combined　with　improved　experiment　design　method,　was　constructed　through　few　times　structure　analysis.　A　second-order　evaluation　function　was　set　up　for　each　constraint　condition.　Rational　move　limits　for　each　design　variable　were　calculated　in　terms　of　the　relative　errors　between　constraint　response　surfaces　and　evaluation　functions.　The　optimization　strategy　formed　with　less　computation　and　higher　precision.　Examples　are　given　to　show　the　efficiency　and　stability　of　this　optimization　strategy.