The digital camera is a powerful tool to capture images for rise in image processing and colour communication. However, the RGB signals generated by a digital camera are device-dependent, i.e., different digital cameras produce different RGB responses for the same scene. Furthermore, they are not colorimetric, i.e,, the output RGB signals no not directly correspond to the device-independent tristimulus values based on the CIE standard colorimetric observer. One approach for deriving a colorimetric mapping between camera RGB signals and CIE tristimulus values uses polynomial modeling and is described here. The least squares fitting technique was used to derive the coefficients of 3 X n polynomial transfer matrices, yielding a modeling accuracy typically averaging 1 DeltaE units in CMC(I:II when a 3 X II matrix is used. Experiments were carried out to investigate the repeatability of the digitizing system, characterization performance when different polynomials were used, modeling accuracy when 8-bit and 12-bit RGB data were used for characterization, and the number of reference samples needed to achieve a reasonable degree of modeling accuracy. Choice of characterization target and media and their effect on metamerism have been examined. It is demonstrated that a model is dependent upon both media and colorant, and applying a model to other media/colorants can lead to serious eye-camera metamerism problems. (C) 2001 John Wiley & Sons, Inc.