Shortage of soil hydraulic parameter data at different scales is a stumbling block for further understanding of hydrologic, hydro-climatic, and general circulation models. Hydrologic modeling requires land surface parameter data such as porosity, residual soil water content, and saturated and relative hydraulic conductivities at multiple scales as inputs. A scaling method using Artificial Neural Networks (ANN's) in conjunction with Bayesian statistics and Markov chain Monte Carlo methods to improve the prediction accuracy of the models is presented. A novel application of ANN's in hydrology, this method handles pedo-transfer functions and scaling of soil hydraulic parameters simultaneously. Multi-scale values for soil hydraulic parameters are derived using remotely sensed land surface data from AQUA and TERRA satellites, SSURGO soil properties database and available point-scale data. The ANN-based scaling process is carried out in multiple steps - from the satellite footprint scale to the field/soil pedon scale, and in turn from field scale to the point scale. This is considered necessary since the same factors do not influence the value of the soil hydraulic parameters at different scales. Thus, suitable changes are made to the network to accommodate this variability. The algorithm is tested at two sites, one each in the Rio Grande basin, New Mexico, and the Southern Great Plains region in Oklahoma. These sites are chosen as they are widely different in landscape characteristics such as soil type, topography, vegetation, and management practices, while ensuring sufficient data availability for validation. The results of the Bayesian training are also compared with those of the conventional training to study the improvements.