The invention discloses a method for automatically removing movement artifacts of near-infrared spectral signals. The method includes steps of firstly, computing movement standard deviation of the near-infrared spectral signals, automatically selecting threshold values and marking movement artifact regions of original signals; secondly, decomposing empirical modes of the near-infrared spectral signals to obtain intrinsic mode functions with clear physiological significances; thirdly, computing suspected movement artifact regions of each intrinsic mode function, combining the suspected movement artifact regions with the movement artifact regions, which are obtained in the first step, of the original signals, processing the different intrinsic mode functions and removing the movement artifacts; fourthly, reconstructing the signals. The method has the advantages that the threshold values are automatically selected according to probability distribution features of the near-infrared spectral signals, and intervals containing the movement artifacts are automatically detected according to the probability distribution features of the near-infrared spectral signals; the empirical modes of the near-infrared spectral signals are decomposed, accordingly, the intrinsic mode functions with the clear physiological significances can be obtained, obviously abnormal intrinsic mode functions can be processed, and the movement artifacts can be removed while physiological information in the NIRS (near-infrared spectral) signals can be effectively reserved.