Is it possible to start with observed time-domain concentrations of final product substances and automatically create both the topology and sizing of the network of chemical reactions? In other words, is it possible to automate the process of reverse engineering a network of chemical reactions? Although it may seem difficult or impossible to automatically infer both the topology and numerical parameters for a network of chemical reactions from observed data, the results presented in this talk seem to answer this question affirmatively. Genetic programming can automatically create complex networks comprising both a graphical structure and numerical values that exhibit prespecified behavior in fields such as analog electrical circuits, controllers, genetic networks, and antennas. In this talk we will present some results obtained by one of the fathers of Genetic programming, John R. Koza, in the field of automatic synthesis of metabolic pathways.