Blood oxygen saturation is an important physiological parameter of human body. The accurate measurement of it is important to both physiological research and medical application. Dual-wavelength method is widely adopted in nonivasive detection of blood oxygen saturation. In this method, the calculation of blood oxygen saturation is based on the identification of pulse waveform and the extraction of peak characteristic values. But the pulse waveform obtained from this method merely has distinctive features such as the QRS complexes in ECG, so the ratio of correct detection of pulse waveform is always low. An autoregressive model of heart-beat intervals is developed. The output of the differential method is compared with that of the AR model so as to distinguish the false or missing detection. The ratio of correct detection is improved by the use of this method.