Abstract: Soil fertility evaluation plays an important guidance role in promoting agricultural reconstruction, improving soil productivity, and recommending proper fertilization. Based on the investigation and evaluation of arable land fertility in Changsha County of Hunan Province, Southcentral China in 2007-2010, this paper analyzed and modeled the determining factors of paddy soil fertility in the doublerice cropping region of Changsha by using the statistical analysis method of classificationregression tree (CRT). A total of 22 independent variables associated with paddy soil fertility, such as site condition, soil physical and chemical properties, and farmland basic facilities were selected, and the ratio of surveyed yield to potential yield for eliminating the rice variety impact was treated as the dependent variable for representing the paddy soil fertility level. Our results suggested that elevation, topographic position, field surface slope, drainage capacity, water transport mode, soil parent material, soil texture, soil plow layer depth, soil organic mater content, and soil available nitrogen concentration were the main factors affecting the fertility characteristics of paddy soil in the double-rice cropping region of Changsha, and these 10 factors could be used to construct a regression tree model to well predict the paddy soil fertility level. Considering its good performance in analyzing complicated relationships of numerous variables and in handling large volume of data set, the CRT method was recommended to be used in the arable land fertility evaluation in other places of China.