Abstract: Abstract: The accurate prediction of the distribution of hilly farmland with purple soil nutrient is of great importance for fertilizing crops reasonably. Also it is very important as a means of increasing farmer income, protecting the environment and source, and improving the sustainable development of agriculture. Many research studies have focused on the spatial variation of soil nutrient, and conclude that the land use type and topographic features of the land are the key factors for the spatial distribution of soil nutrient. However, we can not predict the distribution of hilly farmland with purple soil nutrient based on presently known results. From the knowledge gaps outlined above, the objectives of this study were to predict the regularity of purple soil nutrient for hill farmland. We collected a typical hilly farmland of purple soil (2 km2) in Jiangjin, Chongqing as the study area. One model was built based on terrain attributes, and the other one was built based on a combination of terrain attributes and land use types. The study compared the complexes and fitting for the two models to identify which one is better. The results showed that organic matter and alkali-hydrolyzable nitrogen were positively correlated with topographic factors. This result agrees with previous results that indicated that alkali-hydrolyzable nitrogen is often positively correlated with organic matter in soil because the mineralization of organic matter could generate alkali-hydrolyzable nitrogen. Available phosphorus and rapidly available potassium were weakly correlated with topographic factors, suggesting that terrain attributes have little effect on those two soil nutrients. There were significant differences (P<0.01) of the average of organic matter, alkali-hydrolyzable nitrogen and available phosphorus between crop land and paddy field. But we did not observe a significant difference (P=0.34) of rapidly available potassium in these two fields. This suggests that land use type has a significant effect on the spatial distribution of organic matter, alkali-hydrolyzable nitrogen, and available phosphorus, but except for rapidly available potassium. Thus, if we built the soil nutrient prediction model based only on the land use type, rapidly available potassium may not need to be considered. A comparison of the performance for the two models demonstrates that prediction models based on terrain attributes are convenient, accurate and applicative. This suggests that the predictor variable of land use type as calculated in the model did not positively help to improving the prediction. This study could provide basic scientific evidence for the reasonable fertilization of hilly farmland.