The ability to record different signals when imaging surfaces of differing spectral properties is a pre-requisite for being able to reconstruct spectra from data captured by a color imaging device. The question of which spectral differences can be discriminated between by a given input color imaging device is therefore addressed here by sampling the space of all possible surface spectral reflectances, simulating a device's response to the samples and then identifying regions in which discrimination is possible. Results of applying the technique to the model of an actual and a theoretical, narrow-band camera are also shown to illustrate the information provided by the approach presented here.