In order to use a portable near infrared spectrometer preferably and quickly to conduct the real–time on scene test in tobacco production, the measured spectrum was usually moved to the spectral model already established in laboratory by mathematical method to directly conduct the forecast analysis, which saves cost compared with rebuilding a model. In this paper, slope intercept algorithm (SBC), piecewise direct standardization (PDS) and canonical correlation analysis (CCA) were used and compared in transfer of portable near infrared spectrum of the whole leaf, cigarette silk and powder to the powdered tobacco on Fourier near–infrared spectroscopy model, followed by predicting values of root mean square error (RMSEP). The results showed that RMSEP value of near infrared spectrum of the powder, cigarette silk and whole leaf calibration transferred by SBC algorithm were 0.736 5, 0.798 8, 0.298 6 respectively, by PDS were 0.889 2, 0.640 9, 0.300 7 respectively, by CCA were 0.716 5, 0.761 3, 0.536 9, respectively. RMSEP results after calibration transfer with CCA were better than those with SBC and PDS, and better the RMSEP predicted by rebuilt model using the portable near infrared spectrum directly.