The wavelet transforms are integrated with transform domain LMS algorithm and variable step - size LMS algorithm, from which a new adaptive filtering algorithm is presented based on discrete wavelet transforms ( NDWT - LMS). The algorithm can reduce the cross - correlation of input signals effectively and can overcome the conflict between high convergence rate and low steady - state maladjustment in LMS algorithm which is caused by the fixed step - size factors. The computer simulation results indicate that the higher convergence rate and lower maladjustment noise are obtained by using this algorithm in comparison with LMS algorithm, and the new algorithm is effectively applicable to the adaptive systems.