The O2-PLS method was applied to basic research on non-invasive measurement of human blood glucose. The O2-PLS method can respectively divide the spectral matrix and concentration matrix into aim factors and aimless factors,only using aim factors to build calibration model. Two-component glucose solution,four-component glucose solution and human plasma samples are taken as research objects,and test set validation method is used to evaluate prediction results. The results show that prediction correlation coefficients are all 0.999 and root mean square errors of prediction (RMSEP) were 1.446,1.93,.274 mg/dL respectively. The prediction accuracy of O2-PLS models are higher than that of PLS models. Moreover,the complexity of samples has less effect on the prediction accuracy of O2-PLS models compared with traditional partial least squares (PLS) models. The experimental results indicate that O2-PLS method is applicable to complex samples detection.