The wavelet transform was applied to process the accelerometer signals derived from human walking. The accelerometer signals were first decomposed at different levels utilizing the multi-scale and multi-resolution characteristics of the discrete wavelet transform. After the determination of both the mother wavelet and the optimal decomposition level, human gait series can thus be extracted from the eigen scale of the accelerometer signal. Compared with the method that detects peak values directly from accelerometer signals by thresholding, the wavelet transform gives higher detection rate of peak values on the eigen scale of the accelerometer signals. Even when the accelerometer signals are exposed to serious noise, experimental results still demonstrate that the wavelet approach can guarantee the precision of the extracted gait series, which is of vital importance for the subsequent analyses. It can be concluded that wavelet transform is an effective tool for the extraction of gait rhythmicity. The wavelet transform will be helpful in identifying the characteristics of other physiological signals.