We propose here a new evolutionary algorithm, the RBF-Gene algorithm, to optimize Radial Basis Function Neural Networks. Unlike other works on this subject, our algorithm can evolve both the structure and the numerical parameters of the network: it is able to evolve the number of neurons and their weights. The RBF-Gene algorithm's behavior is shown on a simple toy problem, the 2D sine wave. Results on a classical benchmark are then presented. They show that our algorithm is able to fit the data very well while keeping the structure simple – the solution can be applied generally.