Pulse diagnosis, with a history of more than 2,000 years, is a non-invasive method to measure the holistic health of the human body. To extract spatial features from wrist pulse signal, three-dimensional pulse images (3DPI) was proposed by Luo since 2012. A new problem, 3D image denoising, has arisen in research on pulse signal. Robust principal component analysis (RPCA) was used for 3DPIs denoising, but the result was unsatisfactory for lacking prior knowledge of the signal and noise. In order to achieve better denoising performance and three-dimensional visual performance of 3DPIs, we proposed a novel method, Channel-Weighted Robust Principal Component Analysis (CWRPCA), replacing RPCA from the previous 3DPI preprocessing process. In this research, we make a comparison by three indicators: (1) percentage of valid periods, (2) error be-tween channels, and (3) peak signal-to-noise ratio (PSNR). The experimental results prove that CWRPCA has the most optimal result, evaluated by the three indicators. In conclusion, CWRPCA has better performance in image denoising for 3DPI and helps subsequent analysis in pulse diagnosis research.