In this work, a database containing 1337 data points for selective CO oxidation over Cu based catalysts was constructed from 20 research publications and used for knowledge extraction by artificial neural networks. The experimental CO conversions reported in each publication were successfully predicted by a neural network trained using the data from the remaining 19 publications unless that one publication contained unique variables. The effects and relative significances of the catalyst preparation variables (such as Cu loading, second metal additive, support type, and preparation method) and operating variables (such as reaction temperature, feed composition, and feed flow rate/catalyst weight ratio) were also determined quite successfully by the artificial neural networks. We conclude that neural network modeling can be used to extract valuable experience and knowledge accumulated in the published data and can help researchers to plan new experiments in a more effective manner.