Abstract: A nonlinear filtering method based on principle component analysis (PCA) was proposed according to the statistical characteristics of the Doppler ultrasound blood flow signal and wall thump signal. This method was applied to computer-simulated and clinically collected carotid Doppler signals, and compared with the conventional high-pass filtering method. The mean blood velocity was estimated using the proposed method with a relative error of 5.62%, which demonstrated improved accuracy compared to 12.83% by using the high-pass filtering method. The experiment results indicate that the PCA-based nonlinear filtering method can remove wall thump signal under a broad wall-to-blood ratio (WBR) range while preserving low frequency blood component, thus it can be applied to the separation of the blood and wall components in Doppler ultrasound systems, and facilitate better accuracy and reliability of the further parameter extraction and signal analysis.