Abstract: Transfer model is an approach of fertilizer recommendation that takes plant growth affecting factors into an integrated account and makes all-sided fertilization recommendation. It has proved that transfer model is superior to traditional fertilizer response function in fertilizer recommendation, but its adaptability in different geographic scale is not clear. This paper studied the adaptability of transfer model in different geographic scale by comparing with fertilizer response function by a calibration trial. The study was conducted during 2000-2002 in north Xinjiang, and orthogonal design included experiment scheme was put into practice in 31 dispersed spots each year. In all experiment plots, crop cultivation measures , such as cultivars (cotton, Xinluzao 6) , date of sowing, density of plants, irrigation, pest control and chemical regulation was uniformed strictly to overcome environmental side-affects. Cotton lint yield, soil organic matter content, soil total N content, soil alkali hydrolysable N content, soil available P content and soil available K content were analyzed and taken as site variables to simulate transfer model in different geographic scale. The results indicated that transfer model could be established between site variables and fertilizer response trend coefficients in high statistical critical. Though empirically, these regression parameters are meaningful in fertilization. Compared with traditional fertilizer response function, fertilizer suggestion released by transfer model is significantly negatively correlative to soil alkalihydrolysable N content or soil organic matter content, which suggests that transfer model could adjust N application rate according to soil N supplying capability, especially in small-scaled transfer model. Both large and small geographic-scaled transfer model is feasible in fertilizer prediction. However, the precision of fertilizer predication is higher in small geographic-scaled transfer model than in large geographic-scaled transfer model. Even so, the precision of fertilizer suggestion released by both small and large geographic-scaled transfer model is better than in fertilizer response model.