One of the main challenges in the noninvasive sensing of blood glucose by near-infrared (NIR) spectroscopy is that human temperature changes interfere with the measurement spectrum. In this paper, a spectral pretreatment method based on external parameter orthogonalization (EPO) is proposed to eliminate the spectral variations from temperature interferences caused by the temperature changes at the measured position. This method only needs to collect the diffuse spectrum when the body temperature changes in advance, using whichwe can obtain the filtering matrix to eliminate the temperature interference.This matrix could be used to calibrate the spectrums at different body temperatures to the reference level. This method establishes a model for external disturbance variables separately in advance and separates it fromthe model between glucose concentrations and the diffuse reflectance. The principle of EPO indicated that spectral space is composed of interference signal space and useful signal spacethat are orthogonal to each other. In other words, the temperature response and the glucose concentration response in the spectral is orthogonal to each other. However, in the actual situation, the instrument system drift, common-mode disturbance such as human body sweating often leads to useful signal and interference signal accidentally correlational that does harmto the effectiveness of EPO.Therefore, we first use the differential correction method based on the spectra from the reference position and measuring position on the original spectrum. It has been proved that the differential correction method can eliminate the common-mode interference brought by the instrument system. In addition, the temperature response part and the concentration response part of the obtained absorbance spectrum are orthogonal to each other. In this paper, the spectral data were obtained by Monte Carlo simulation of human three-layer skin model, and the parameters of the simulated samples were set according to the actual human experiment. The EPO method was used tocalibrate temperature interference in spectra after the differential correction method, and then a partial least squares regression (PLSR) model was established based on the corrected spectral data. Compared with the spectral modeling results before calibration, the root means square error of calibration (RMSEC) was reduced; the correlation coefficient was improved, and the number of principal components was reduced, which indicated the effectiveness of the temperature correction method-EPO.