This paper deals with the application of neural network technique for the three-dimension interpretation of mise-à-la-masse data from the Takigami geothermal field in Kyushu, which is one of the most active geothermal area in Japan. To understand the structure of the geothermal field, a 4-layers neural network had been developed. The training algorithm for the network is back-propagation with five paradigms, e.g. on-line back-propagation, batch back-propagation, delta-bar-delta, resilient propagation (RPROP) and quick propagation, were applied to find out the most efficient one. The network was trained with 3-D mise-à-la-masse simulation data set, including 864 cases of a single anomalous resistivity block of 10 Ohm.m moving in the model mesh with background resistivity of 100 Ohm.m. To generate the training data set, a high accuracy algorithm for 3-D numerical simulation, based on finite difference method and the algorithm of the singularity removal, was used. The trained network was tested by a synthetic data and then applied for the real field data set of the study area. The obtained results are remarkably correlated with the other available data from the field such as previous geoelectrical data, formation temperatures, lost circulation zones, hence, promising zones for production or re-injection can be indicated quickly at site of Takigami geothermal field.