The artificial neural networks (ANNs) combining gas sensor arrays possesses several advantages over conventional signal processing in terms of adaptabil-ity (learning, self-organisation, generation and training), noise tolerance, fault tolerance, distributed associated memory, inherent parallelism generating a high speed of operation subsequent to training, and in addition, ANNs are amenable to VLSI implementation, which is great commercial interests of industry. The disadvantages of ANNs stem from the fact that the optimum network topology, data pre-processing and training are very problem dependent, and yet can have a strong influence on the final performance of the networks themselves.It was suggested by Lippman'that a three-layer network have sufficient computational degrees of freedom to solve any problem. Over the passed few years, the three-layer network ANNs have been successfully used with the development of new learning paradigms, in particular, the error back-propagation, and the faster personal computers. In this work, we investigated the application of the ANNs to a gas sensor arrays. Quantification of individ-ual H2S and NO2 components in the gas mixtures were demonstrated using ANNs based on a properly chosen pre-processing function.