Thermal conductivity of ground has a great influence on the performance of Ground Heat Exchangers (GHEs). In general, the ground thermal conductivity significantly depends on the density (or porosity) and the moisture content since they are decisive factors that determine the interface area between soil particles which is available for heat transfer. In this study, a large number of thermal conductivity experiments were conducted for soils of varying porosity and moisture content, and a database of thermal properties for the weathered granite soils was set up. Based on the database, a 3D Curved Surface Model and an Artificial Neural Network Model (ANNM) were proposed for estimating the thermal conductivity. The new models were validated by comparing predictions by the models with new thermal conductivity data, which had not been used in developing the models. As for the 3D CSM, the normalized average values of training and test data were 1.079 and 1.061 with variations of 0.158 and 0.148, respectively. The predictions became somewhat unreliable in a low range of thermal conductivity values in considering the distribution pattern. As for the ANNM, the 'Logsig-Tansig' transfer function combination with nine neurons gave the most accurate estimates. The normalized average values of training data and test data were 1.006 and 0.954 with variations of 0.026 and 0.098, respectively. It can be concluded that the ANNM gives much better results than the 3D CSM.