The use of near-infrared spectroscopy for the monitoring of blood glucose concentration is limited by many ambiguous factors, which leads to the prediction precision is not satisfied. Due to the weak interested signal and the difficulty to quantify the physiological noise directly, the absorbance induced by glucose concentration and temperature was analyzed based on Beer-Lambert Law and displacement between glucose and water. Then the transmittance of glucose aqueous solution in different temperatures was measured by spectrometer to investigate the influence of glucose concentration and temperature. As it's difficult to distinguish the influence of temperature from the diffuse reflectance, the Monte Carlo simulation was used to compute the light intensity induced by the change in glucose concentration and physiological temperature. Finally, the influence of actual physiological temperature on the prediction model of glucose concentration was estimated based on the oral glucose tolerance tests of two diabetics. The result showed that, near the normal physiological temperature, the intensity of diffuse reflectance caused by -0.1 °C change in temperature was equivalent to that caused by 2.7 mmol/L change in glucose concentration. Moreover, the proportion of prediction error induced by temperature to the total error was more than 50%.