Field water retention capacities (θ_{-30 kPa}) and wilting coefficients (θ_{-1 500 kPa}) of ninety soil samples in the Dagu River Basin were predicted separately with four PTF_S, i.e. point regression method, linear regression method, nonlinear regression method and artificial neural network method, and their spatial variabilities were analyzed with the aid of traditional statistic and geostatistic methods. The traditional statistics revealed that the nonlinear regression method was the best with the variation coefficients of θ_{-30 kPa} of all the soil samples, being always less than θ_{-1 500 kPa}, however, no matter measured or predicted values, both belonged to the category of moderate in spatial variability. The geostatistics also showed that both measured and predicted θ_{-30 kPa} and θ_{-1 500 kPa} demonstrated varied nugget effects, moreover, θ_{-30 kPa} always had stronger spatial dependence than θ_{-1 500 kPa} did. Analysis of the parameters of semivariance model for θ_{-30 kPa} and θ_{-1 500 kPa} ultimately revealed that the artificial neural network model could most truthfully characterize spatial variability of the soil water retention capability in the experimental zone.