© 2015 World Scientific Publishing Company. The monitoring of arterial oxygen saturation (SaO2) is a common practice in both clinical and home environments, and the process of monitoring can be exercised invasively or non-invasively. In the past decades, the pulse oximeter is one of the most popular non-invasive devices that use photoplethysmography (PPG) to measure SaO2. As the pulse oximeter requires calibration prior to application in clinical practice, a significant number of calibration methods have been proposed based on experimental data collected from human volunteers. Alternatively, models may be derived from the Lambert-Beer law or the photon diffusion equation (PDE). However, most of such calibrated oximeter can only provide accurate readings of SaO2 at high versus the low levels. We propose to apply an extra-boundary condition to solve the PDE, and then to develop a model-based calibration method that relate optical measurements to the level of SaO2 in this work. Then, we validate our method against previous model-based methods and an oximeter simulator with higher accuracy when the level of SaO2 is greater than 90%. In practice, our model-based method can still maintain a good performance when the level of SaO2 decreases to 60%, thereby demonstrating high potential in the accurate evaluation of the oxygen level of patients by PPG.