Device independent color requires that all color imaging devices 'speak' in CIE colorimetric terms. A color image scanner is one component of the desktop color imaging chain that needs to be colorimetric if color is to be device independent. Most commercial color scanners are not colorimetric. Instead, they are built as color densitometers, emulating commercial graphic arts scanners. There are a wide variety of spectral response options if the design goal is a colorimetric scanner. This paper explores 'almost' color mixture spectral response options using Neugebauer's Colorimetric Quality Factor (CQF) as a design merit function. A set of empirical equations relating the average CIEL*a*b* color difference to the CQF under three CIE illuminants (A, F2 and D65) is developed, based on a database of over 1000 colors representing a wide variety of colorants and color imaging technologies. Our goal is to empirically bridge this gap between the CQF specification for each channel and the scanner's colorimetric performance, in terms of CIEL*a*b* color difference. For spectral response functions that have a high CQF, a 'universal' transformation matrix is described for transforming 'almost' colorimetric scanner RGB data to approximate CIE XYZ tristimulus data. Scanner peak wavelength and bandwidth requirements for colorimetric and densitometric applications are also reported.