A number of different methods exist for the color characterization of imaging devices such as digital camera systems. In this study, the use of high-order polynomials and artificial neural networks for color camera characterization are compared and contrasted. A quantitative evaluation of their performance is determined for a typical commercial camera system. The importance of independent training and testing sets is stressed and the effect of the number of samples in the training set is evaluated. The results show that, if the best performance is considered, the two models are approximately comparable. Any performance advantage obtained from using a neural network for device characterization does not seem to be warranted given the additional risks of using such systems. The effect of training set size seems surprisingly small for both polynomial and neural systems with generalization performance only being seriously affected for training set sizes less than about 100.