Gas sweetening is a fundamental step in gas treatment processes. In gas sweetening units, acid gases (H2S and CO2) are chemically absorbed from a gas using aqueous alkanolamine solutions, to produce a "sweet gas." The solvent is regenerated in a desorption column and the purified (or "lean") solvent is recycled to the absorption column. Gas sweetening units can be controlled if all of the operation data, for example, sweet gas, lean amine, and rich amine flow rates, concentration, and temperatures existed. In this article, a new method based on artificial neural network for prediction of lean amine concentration is presented. H2S, H2O, CO2, and diethanolamine mole fractions in sour gas and sweet gas have been input as variables of the network and have been set as network output. Among the 130 data set, 92 data have been implemented to find the best artificial neural network structure as train data. Moreover, 19 data have been used to check the generalization capability of the trained artificial neural network named validation data and 19 data have been used to test an optimized network as test data. These predictions can prevent operation problems. The results, according to R value and mean squared error, show good accuracy of this type of modeling.