Abstract: In soil science, the studies of quantitative soil-landscape models have become very active in recent years. However, transportability of these models to unsampled landscapes is still unknown. In this study, a 2 km2 typical hilly study area in southwestern China was chosen as the origin spot. And the soil-landscape model in origin area was built by using multiple stepwise regression with the data of soil properties and terrain attributes. Based on the distance from the origin spot, the study area was divided into 4 regions (denoted A, B, C, D area). All these 4regions which had similar terrain and soil environment were used to analyze the transportability of the soil-landscape model. The results showed that: soil-landscape model got predictions with high accuracy level in general. As the distance from origin spot increases, the mean absolute error (MEA) and root mean of squared error (RMSE) had shown a tendency to increase as well. These meant the accuracy of prediction was reduced with the distance. However, when the distance reached to a threshold the accuracy was no lower any more. In 4 regions A, B, C and D, there were proportions that the relative error (RE) was more than 20%. And the farther from origin area, the greater they became. Meanwhile the proportion that relative error (RE) was less than 10% decreased gradually.