An artificial neural network has been trained using a data set that was obtained for the corrosion of steel in seawater. The
model appears to fit the data relatively well, in that the rms (root mean square) residuals for the training, validation and
independent test sets are all relatively low. However, several of the apparent effects of the various input parameters are
counter to normal expectations. While some of these effects may be genuine, the training data has a strong structure, with
a high degree of correlation between several of the input parameters. Consequently it is possible that the observed effects
are artifacts that result from the rather limited coverage of the problem domain by the training data. Better methods are
required for the validation of neural network models of corrosion processes.