The traditional method of measuring arterial oxygen saturation is that R value, the ratio of alternating component of the logarithmic photoplethysmography, is firstly computed and then the linear regression model is established by experiment. The R value computation is a dimension reduction process based on Lambert-beer law, which aims at eliminating the influence of optical path andminimizing the impact of individual differences. When taking scattering into consideration, the dimension reduction process loses information, introduces the system error and limits the precision of measurement. In order to reduce the measurement error resulting from the scattering effects, this paper presents a new method that the peak and valley values of dual-wavelength logarithmic photoplethysmography waves are used as the independent variables to develop a linear regression model to predict the arterial oxygen saturation. During the experiment, the in-vivo measurements were carried out on 23 healthy volunteer and 133 samples of photoplethysmography waves and the reference value of oxygen saturation were recorded. To compare the predictive performance between the new method and the R value method, 90 samples were randomly selected as modeling sets and the remaining 43 samples were used as prediction sets. Random selection of modeling sets and prediction are executed 10 times. The average related coefficients of the prediction sets of the new method and the R value method are 0.890 6 and 0.846 8, and the average root mean square errors are 0.889 6% and 1.037 3% respectively. Results indicate that the performance of the new method is better than the one of the R value method, and the predictivemodel based on 4 parameters can improve the stability and accuracy of measurement. And the new method has guiding significance to the measurement of human body’s blood physiological information based on limited wavelength spectrum data.