The reference-wavelength method is proposed to diminish the influence of noises on glucose measurement by differentially processing two signals at the reference and measuring wavelengths. At the reference wavelength, the radiation intensity is insensitive to the changes of glucose concentration. Therefore, it can be used as the internal reference to estimate the noise and then to extract the effective glucose signal at the other wavelengths. The validation experiments are constructed in the non-scattering samples with the reference wavelength of glucose at 1525 nm. The results show that the reference-wavelength-based glucose-specific signal extracting method can largely improve the glucose prediction precision from 17.56 to 8.87 mg/dL in the two-component experiment and from 26.82 to 9.94 mg/dL in the three-component experiment.