A typical peak-cluster depression area was selected as the study area. Based on the relationships between total soil nitrogen content in 0~20cm depth and terrain attributes by multiple linear stepwise regression method, the spatial variability of total nitrogen was studied by ordinary Kriging (OK), univariate Cokriging (COK) and multivariable Cokriging (MCOK) methods. Results showed that the spatial distribution of total nitrogen can be described by two regression models. Distance from ridge of mountains (VDFR) entered into the regression equation for the model (1), and distance from ridge of mountains (VDFR) and relief degree of land surface (VRS) entered into the regression equation for the model (2). Therefore, VDFR was selected as auxiliary variable to predict total nitrogen by COK method, and VDFR and VRS were selected as auxiliary variables to predict total nitrogen by MCOK method. The Kriging interpolation indicated that when the correlation coefficient between total nitrogen and terrain attributes was small, the accuracy of COK was increased litter. While MCOK can promote the accuracy notable. This study can provide a new idea and method for evaluation and ecological reconstruction of the karst rock desertification.