An improved spectrum extraction method is proposed based on the dynamic spectrum theory to improve the precision. First, the drift baseline is fitted to restrain the effect of relative motion between the finger and the probe. Then, the spectrum is extracted using the square-sum-root method as a replacement of the peak-peak value method to reduce the impact of system random error. Taking the measurement of oxygen saturation for instance, uncertainty analysis is performed for the oxygen saturation coefficient Q. Compared with the peak-peak value method, the square-sum-root method can reduce the uncertainty of oxygen saturation coefficient Q to 38% of the original value. Ten data sets were sampled from ten volunteers, and the uncertainty of the coefficient Q obtained using the two methods was directly evaluated and compared. The average ratio of the Q uncertainty by the two methods turned to be 42%. Both the theoretical analysis and the experimental results show that the square-sum-root method proposed can be used to extract the effective spectra more accurately from the photoplethysmography, suggesting a meaningful progress in noninvasive blood components measurement.