Scaling properties of neural networks, i.e. relations between the number of hidden units and the training or generalization error, recently have been investigated theoretically with encouraging results. In this paper we investigate experimentally, whether the theoretic results may be expected in practical applications. We investigate different neural network structures with varying number of hidden units for solving two time series prediction tasks. The results show a considerable difference of the scaling behavior of multilayer perceptrons and radial basis function networks.