In this paper, we suggest a new approach for reverse engineering gene regulatory networks, which consists of using a reconstruction process that is similar to the evolutionary process that created these networks. The aim is to integrate prior knowledge into the reverse engineering procedure, thus biasing the search towards biologically plausible solutions. To this end, we propose an evolutionary method that abstracts and mimics the natural evolution of gene regulatory networks. Our method can be used with a wide range of nonlinear dynamical models. This allows us to explore novel model types such as the log-sigmoid model introduced here. To allow direct comparison with other methods, we use a benchmark dataset from an in vivo synthetic-biology gene network, which has been published as a reverse engineering challenge for the second DREAM conference.