In　this　paper,　a　fast-search　algorithm　is　introduced　to　reduce　the　complexity　of　LSF　quantization　in　speech　coding.　A　new　inequality　between　the　weighted　mean　and　the　weighted　Euclidean　distance　is　derived.　Using　this　inequality,　many　codewords　that　are　impossible　to　be　the　nearest　codeword　are　rejected　directly.　The　proposed　algorithm　produces　the　same　output　as　conventional　full　search　algorithm　and　the　experiment　results　confirm　its　effectiveness.　©　2006　IEEE.