Model transfer is a key common technical problem in the near infrared spectral analysis technology. By seeking feasible mathematical methods between the two instruments that have the same working principle, we can make the model which was set up on one instrument be applied to another one. In this paper, with 150 flue-cured tobaccos as test samples, with two Bruker MPA near infrared spectrometer and one Thermo Antaris near infrared spectrometer as the research object. We obtained spectral data by integrating sphere diffuse reflectance. Processed and analyzed the spectral data by using the first derivative and standard normal variate (SNV) transformation, and calculated the value of residual error between different instruments, first moment, signal probability density and maximum signal to noise ratio(SNR) and so on. Also, we established mathematical model of total sugar content by partial least squares (PLS) to test the effect of model transfer. The results showed that first derivative could reduce the first moment and transfer the deviation between different instruments into the standard Gaussian distribution, but at the same time, it could also put SNR down. SNV could also reduce the first moment and even could do better than first derivative, and it could increase SNR significantly, but SNR could not transfer the deviation into standard Gaussian distribution, which would need other ways to make up for it. The combination method of the first derivative and SNV can retain the advantages of both, and make up for the disadvantages of individual treatment, and it can settle the model transfer problem caused by different instrument types and different using time between different instruments that work in the principle of Fourier, which is based on the integrating sphere diffuse reflectance. This method is an ideal model transfer method without the prototype.