An overview is given of the current usage of artificial neural network
s as mathematical models for non-linear calibration procedures. The em
phasis is on practical aspects: the choice of the calibration samples,
the required network characteristics for a given problem, various tra
ining methods and their efficiency and the validation of the network m
odels. Some problems with the application of neural networks in multiv
ariate calibration are considered, together with recent research aimed
at solving these problems.