Abstract: A denoising method based on the soft-thresholding wavelet frames is proposed for Doppler ultrasound signals from the bi-directional blood flow. A direction separation step is firstly carried out to avoid the direction distortion of Doppler signals, which is induced from the non-linear soft-thresholding processing. Then a pair of separated signals are denoised independently. Finally Doppler signals from the bi-directional blood flow are reconstructed from the denoised separated signals. This approach has been applied to the simulated Doppler signals from femoral arteries. The experiment results show that the signal-to-noise-ratio (SNR) of denoised signals is enhanced without affecting the information of the bi-directional blood flow. The estimation precision of the maximum frequency and mean frequency are also improved. It is shown that this method is valid to denoise Doppler ultrasound signals from the bi-directional blood flow.