Thermal infrared spectra of glasses and phyllosilicates have similar shapes leading to a proposed ambiguity in deconvolution results. We quantify the spectral separability of these two classes using discriminant analysis and linear deconvolution. We apply the deconvolution technique to single spectra, and two- and four-component mixtures. Missing end-members result in quantifiable uncertainties between modeled abundances of glasses and phyllosilicates. Mixtures containing silica-K2O glass are susceptible to overestimating phyllosilicates with minimal increases in root-mean-square error if the silica-K2O glass spectrum is not present in the end-member set. Missing phyllosilicate end-members are likely to be modeled as combinations of other phyllosilicates rather than glasses, with saponite, nontronite, and halloysite showing the highest uncertainties. Empirically, ¿15% of the total glass plus phyllosilicate abundance incorporates all uncertainties in derived glass abundance if the glass in the mixture is in the end-member set. If the glass in the mixture is excluded from the end-member set, these uncertainties increase to ¿65%. Four-component mixtures show that glasses and phyllosilicates act as a two-component mixing system but can induce small uncertainties in other components. Applying the maximum derived uncertainties to deconvolutions of surface type 2, measured by the Mars Global Surveyor Thermal Emission Spectrometer, we find that silica-K2O glass abundance is above the instrument detection limit and a likely component of the surface. Pure silica glass, a proposed amorphous weathering product, is the least likely candidate for confusion with phyllosilicates and its noninclusion in models of Martian spectra suggests it likely is not a component of the Martian surface.