Several of the icy satellites of Saturn show the spectroscopic signature of the asymmetric stretching mode of C-O in carbon dioxide (CO2) at or near the nominal solid-phase laboratory wavelength of 4.2675 ??m (2343.3 cm-1), discovered with the Visible-Infrared Mapping Spectrometer (VIMS) on the Cassini spacecraft. We report here on an analysis of the variation in wavelength and width of the CO2 absorption band in the spectra of Phoebe, Iapetus, Hyperion, and Dione. Comparisons are made to laboratory spectra of pure CO2, CO2 clathrates, ternary mixtures of CO2 with other volatiles, implanted and adsorbed CO2 in non-volatile materials, and ab initio theoretical calculations of CO2 * nH2O. At the wavelength resolution of VIMS, the CO2 on Phoebe is indistinguishable from pure CO2 ice (each molecule's nearby neighbors are also CO2) or type II clathrate of CO2 in H2O. In contrast, the CO2 band on Iapetus, Hyperion, and Dione is shifted to shorter wavelengths (typically ???4.255 ??m (???2350.2 cm-1)) and broadened. These wavelengths are characteristic of complexes of CO2 with different near-neighbor molecules that are encountered in other volatile mixtures such as with H2O and CH3OH, and non-volatile host materials like silicates, some clays, and zeolites. We suggest that Phoebe's CO2 is native to the body as part of the initial inventory of condensates and now exposed on the surface, while CO2 on the other three satellites results at least in part from particle or UV irradiation of native H2O plus a source of C, implantation or accretion from external sources, or redistribution of native CO2 from the interior. The analysis presented here depends on an accurate VIMS wavelength scale. In preparation for this work, the baseline wavelength calibration for the Cassini VIMS was found to be distorted around 4.3 ??m, apparently as a consequence of telluric CO2 gas absorption in the pre-launch calibration. The effect can be reproduced by convolving a sequence of model detector response profiles with a deep atmospheric CO2 absorption profile, producing distorted detector profile shapes and shifted central positions. In a laboratory blackbody spectrum used for radiance calibration, close examination of the CO2 absorption profile shows a similar deviation from that expected from a model. These modeled effects appear to be sufficient to explain the distortion in the existing wavelength calibration now in use. A modification to the wavelength calibration for 13 adjacent bands is provided. The affected channels span about 0.2 ??m centered on 4.28 ??m. The maximum wavelength change is about 10 nm toward longer wavelength. This adjustment has implications for interpretation of some of the spectral features observed in the affected wavelength interval, such as from CO2, as discussed in this paper.