Prediction of gas thermal conductivity is crucial in the heat transfer process. In this article, we develop a novel method to estimate conductivities of binary gaseous mixtures at atmospheric pressure. The method is a neural network scheme consisting of two consecutive multilayer perceptrons (MLPs). The first MLP estimates pure component conductivities as a function of critical temperature, critical pressure, molecular weight, and temperature. The conductivities calculated in the first MLP as well as molecular weights of both compounds and mole fraction of the light components are fed to the second MLP to predict the thermal conductivity of the mixture. The proposed model was trained and tested through a large set of experimental data over wide ranges of temperatures, compositions, and substances. Comparing the test and training results indicates that the accuracy of the neural model is remarkably better than other alternative methods proposed in the literature. Conventional conductivity correlations require more input parameters which are not available for many gases. Also, correlations recommended for pure gas conductivity are usually valid for a particular range of temperature and substances. However, the MLP scheme is able to cover a wide range of temperatures and substances with a few numbers of parameters which are abundant for most gases.