A common method for obtaining scene-referred colorimetry estimates is to apply matrices to radiometrically linearized capture device signals, obtained either from digital cameras or scans of photographic film. These matrices can be determined using different test colors and different error minimization criteria. Since the spectral sensitivities of these capture media typically do not meet the Luther condition, the application of the matrices warps the spectral locus, as analyzed by the capture devices and media. The warped loci of spectral colors represent the boundary of the gamut of possible scene colors that can be estimated by the device. These gamuts tend to have roughly similar shapes for many popular capture devices and media, and often extend outside the true spectral locus in some areas and do not reach it in others. This observation leads to the conclusion that scene-referred color encoding primaries should be selected based on their coverage of the possible scene colors as analyzed, rather than on their coverage of the gamut of colors seen by the eye. The fact that the analysis gamuts extend significantly beyond the spectral locus in some cases also demonstrates the need for color processing systems to deal appropriately with such colors.