The near infrared quantitative model established through the small test preparation process is difficult to apply in the pilot or large scale production process directly,which discourage the development of near infrared spectroscopy in the pharmaceutical preparation process.To solve this problem,the fluidized bed granulation process with medicinal dextrin was adopted in this paper.Meanwhile,the NIR spectrum data and the moisture contents of samples in the above preparation process experienced both the small test and the pilot test were collected.Then the near infrared quantitative model for moisture content was established with 3 batches of samples in small test.Based on this,a model transfer method,directed direct orthogonal signal correction combined with slope/bias correction(DDOSC-SBC) was proposed to predict the moisture content of pilot samples.The relative standard errors of prediction for two separate pilot test sets were decreased from 51.04% and 26.64% to 4.90% and 3.99%,respectively after using this directed model-transfer method.Compared with undirected direct orthogonal projection technique combined with slope/bias correction(DOSC-SBC) and the model update,this method could remove the interference spectrum information from pilot test samples more effectively.Consequently,this method provides a new idea for application of the near infrared quantitative model transitting from small test to pilot test or even expanding to the large scale production process.