Abstract: The use of satellites with the objective to supply profiles of high resolution to study the atmosphere and together GPS (Global Positioning System) data, opens perspectives to improve the research on climate and the capacity on weather forecast. In this sense, many techniques were developed for retrieving atmospheric profiles (temperature, pressure and water vapor) using GPS radio occultation. A new method based on artificial neural network (ANN) to retrieval temperature profiles is presented in this paper. This technique establishes a non-linear relation with the meteorological variables. Month, latitude, altitude and bending angle were chosen as the input vectors and temperature as the output vector. The satellite CHAMP data was from December 17 (2002) up to February 15 (2003) for a South America region (35S to 5S and 80W to 50W). By comparing the retrieved profiles with the corresponding ones from the CHAMP-ISCD (Challenging Mini-satellite satellite Payload for Geoscientific Research and Application), it can be concluded the ANN is convenient and an accurate tool to get temperature profiles. Its results can be employed to the atmospheric data assimilation to improve the initial condition of the models of Numerical Weather Prediction (NWP).