A mathematical structure that attempts to replicate the process by which the brain learns. The structure is conceived as consisting of a series of layers, the values of statistics computed in one layer being passed to processes (‘neurons’) in the next layer. The aim is that the output layer will provide accurate information about the process under investigation. The layers between the input and the output are called hidden layers. In order to be effective a neural network must be ‘trained’—this is analogous to the estimation of the parameters of a model. The data used for this purpose are referred to as the training sample. Much of the process of statistical modelling can be viewed as the construction of a neural net.