Variably-saturated models require accurate estimates of soil hydraulic characteristics that describe its retention and conductive properties. Ideally, one would prefer to measure required soil hydraulic parameters either in the field or laboratory so that the spatial and temporal variability is sufficiently characterized. However, this is rarely done because (a) a measurement of soil hydraulic properties in the field/laboratory requires significant financial and time investments and (b) the spatial variability of soil hydraulic properties and their scale-dependency makes such characterization rather difficult. In the last decade, researchers have shown a keen interest in developing a class of indirect methods, referred to as PedoTransfer Functions (PTFs)to address this issue. Existing PTF models vary from simple lookup tables that provide hydraulic parameters for particular textural classes, to linear and non-linear regression-based equations, to models that incorporate various physical relationships. One of the more commonly used PTFs is based on the pattern recognition tool called Artificial Neural Networks (ANN)such as the Rosetta program. While ANN-based PTFs have been relatively successful, there are weaknesses. There is still a need for a better pattern recognition tool to improve the PTF's accuracy and reliability. In this presentation, we describe a new promising methodology called Support Vector Machines (SVMs). Unlike the ANNs where the complexity of the ANN structure is fixed apriori and only the prediction error can be minimized, SVMs represent a pattern recognition approach where the overall prediction error and the complexity of the SVM structure are minimized simultaneously. We used the UNSODA database that has been widely utilized for estimating PTFs. The SVM-based PTFs show an improved estimation of soil hydraulic parameters compared to previous approaches (including ROSETTA). All soil hydraulic parameters estimated using SVM-based PTFs showed an improved confidence in the estimates.