Photoplethysmography can be used to noninvasively detect oxygen saturation of human. When detecting by photoplethysmography, because of the disturbance of random noise in the process of signal acquisition, there is high-frequency noise, which affects the final prediction accuracy of oxygen saturation. Therefore empirical mode decomposition(EMD) method based on consecutive mean square error(CMSE) criterion is employed, which can remove high-frequency noise from pulse wave signal. The present paper used a self-developed photoplethysmography acquiring device to obtain the pulse wave signal, employed the above mentioned method to remove high-frequency noise, and adopted frequency spectrum of the signal to evaluate the effect. The results showed that: this method could effectively remove high-frequency noise from pulse wave signal. This would be beneficial for improving the prediction accuracy of oxygen saturation of human.