Abstract:  Algebraic codebook is a popular fixed codebook structure; its searching method directly influences decoding speech quality and computational complexity. The paper puts forward a kind of algebraic codebook GSOS (Graded and Subsection Optimization Search) Method. The method fuses the advantages of pulse replacement, subsection searching and grade optimization method. It first creates good initial codebooks; then proceeds codebook searching by subsection for replacing multiplication by the operation of addition; at the same time,the pulses are graded according to contribution degrees for improving the efficiencies of pulses replacement optimization .By using the proposed method to fixed codebook searching stage of AMR-WB (Adaptive Multi-Rate Wideband Speech Codec),the experiment results showed that the proposed method can make the computational complexities of fixed codebook searching down to 13.75% of depth -first tree search procedure, but the decoding speech qualities were only reduced by 4.01% .The subjective auditory almost did not feel the difference.