The modal mineralogies of 20 mafic and ultramafic igneous rocks were determined from their thermal infrared emission spectra using a linear deconvolution approach, which uses a library of end-member mineral spectra to model a bulk rock spectrum. Over 90% of the modes obtained from thermal emission spectra agree with modes obtained by traditional optical analyses to within the stated error of the optical analyses (5--15 vol%). Library spectra of several compositions within a solid solution series (e.g., plagioclase feldspars labradorite and bytownite) were commonly used in each best fit model and are assumed to represent, in combination, a composition in the rock for which an identical mineral spectrum was unavailable in the library. The accuracy of this assumption was evaluated by calculating a weighted average solid solution compositions for plagioclase and/or pyroxenes from the library minerals used in the model best fit of 14 rocks and comparing these compositions to the actual chemistries measured by electron microprobe or the optically estimated An#. The derived solid solution compositions are generally within 10--15 An# (or Mg#) of the measured composition. Modal data derived from the spectra via the deconvolution were summed and converted to weight percent (wt %) oxides for comparison to standard bulk chemistry data. SiO2, Al2O3, and Na2O+K2O were generally slightly overestimated, and FeO and MgO were typically slightly underestimated. Modes and wt % oxides were plotted on standard rock type classification diagrams and provide broadly accurate classifications, demonstrating that the linear deconvolution technique is successful at deriving useful mineralogical information from thermal infrared emission spectra of bulk rock samples. Furthermore, convolution of the laboratory data to spectral resolutions comparable to remote sensing instruments (specifically, the Mars Global Surveyor Thermal Emission Spectrometer) demonstrates only a ~5 vol % increase in modal uncertainty for minerals present in abundances ≥10 vol %. This study thus provides a strong foundation for the application of this technique to mafic igneous sample spectra from both laboratory and remote sensing instruments. ¿ 2000 American Geophysical Union