Abstract: The study examined the performances of two interpolation methods which allow to account for auxiliary data: co-Kriging, regression-Kriging and tested against ordinary Kriging, to improve the interpolation of soil salinity. The prediction accuracy for the three methods was evaluated in the different sampling densities of the variable of interest by comparison with another group of 80 validation sample points. Results show that whatever the sample size of target variable decreased, co-Kriging and regression-Kriging performed better than ordinary Kriging using auxiliary variables. Moreover, regression-Kriging performed on average more accurate predictions than co-Kriging. The results of the T-test of interpolation error for ordinary Kriging, ordinary co-Kriging and regression-Kriging with different sample sizes indicate that regression-Kriging has the lowest interpolation error than ordinary Kriging and co-Kriging and significant reduction of the interpolation errors is achieved. So, regression-Kriging shows promise for predicting the subsequent soil properties form previously temporal data, or for predicting sparsely located soil properties from dense observations. Moreover, in regression-Kriging, the regression model can be more flexible, such as generalized linear models or non-linear models, which provides a possibility to include more ancillary variables.