The spectral subtraction approach with different flow blood volume in near infrared noninvasive biochemical sensing can eliminate human tissue background interference, but at the same time the pathlength of subtracted spectrum is unknown. Thus, the model will be insufficiently effective if established directly with pathlength-unknown spectra, which is a big problem for subtracted blood volume spectrometry. In the present paper, a simulated experiment was designed to simulate this issue. The orthogonal signal correction method was proposed to eliminate the influence brought by pathlength. Compared with the PLS model before and after orthogonal signal correction, the root mean square error of cross validation (RMSECV) was from 90.17 mg·dL-1 down to 31.62 mg·dL-1, and the correlation coefficient was improved from 0.978 7 to 0.996 8. The experimental results show that using the orthogonal signal correction method can effectively restrain the interference information of pathlength, and improve the prediction precision of calibration model. The conclusion in this paper sets the stage for the practical application of the spectral subtraction approach with different flow blood volume.