Demonstrates how Bayesian methods allow complex neural network models to be used without the overfitting that can occur with traditional training methods. The text presents an insight into complex Baynesian models by a theoretical investigation of priors over functions that underlie them.