Two alternative systems for spectral reconstruction of scene reflectances are compared. Both approaches utilize a CCD camera modulated by a set of filters after which a system model is applied to derive a pixel-by-pixel estimation of scene reflectance spectra. The qualitative difference between the filter sets for the pair of approaches warrants the current comparison. The first approach, described in terms of spectral band pass, uses a set of minimally redundant filters extreme in contrast which pass highly localized radiation. The second approach is based on a set of shaped non-selective filters. Theoretically, an ideal implementation of the first, narrow-band, approach should deliver accurate reflectance estimation for any scene object, independent of the characterization scheme used to build the system model. The second, wide-band, approach has theoretical dependence on the relationship between scene colorants and system model assumptions. When a priori knowledge of the expected colorant makeup of a scene is used to steer the choice of color samples from which the system model is derived, the expected robustness of the wide-band approach is theoretically improved. Results from implementations of the two approaches are contrasted and differences between theoretical expectations and those obtained are discussed.