There are a number of factors other than soil moisture which influence the intensity of the emission from the soil. These include, among others, surface roughness, vegetation cover and texture. For this reason, soil moisture retrieval is usually described as a non-linear and ill-posed problem. An Artificial Neural Network (ANN) has been proven as a method to address this type of problem. Since an ANN is a data driven model, proper input selection is a crucial step in its implementation. The presence of redundant inputs in ANN modeling severely impairs the ability of the network to learn the target patterns. In this paper, the input parameter selections are based on the use of incremental contributions of the variables towards soil moisture retrieval. Field experiment data obtained during the National Airborne Field Experiment 2005 (NAFE’05) is used. The ANN model with a minimum number of relevance input parameters was evaluated against its transferability by testing the trained ANN on a new site. It was found that the Root Mean Square Error (RMSE) obtained is between 5 to 9%v/v which is higher than the globally acceptable soil moisture retrieval accuracy. Results from the study indicate the transferability of ANN is challenging but achievable.