Abstract: The FIR filter exists the defects which pass band or reject band character is bad and filter order is high.The defects make the speech recognition system performance become bad when FIR filter is used in front-end processing.Aiming at the problem，the paper uses Laguerre filter bank to replace FIR filter bank used in front-end processing of ZCPA（Zero-Crossings with Peak Amplitudes） feature extraction.After researching carefully FIR filter parameters determining method，the paper describes Laguerre filter’s parameters determining method，and presents the computing results.The speech recognition experimental results of isolated words and nonspecific persons show that the speech recognition system using Laguerre filter has better robustness to noise than using FIR filter，and the order of Laguerre filter is much lower than one of FIR filter.