One of the major sources of uncertainty for thermobarometry in pelitic rocks is the inability of solution models to represent accurately the non-ideal behavior of important pelitic minerals, in particular garnet, biotite, and plagioclase. In recent years, a large number of different solution models have been proposed for these phases. The purpose here is to test the effects on the overall uncertainty of the most commonly-used solution models using the garnet-muscovite-annite-plagioclase (GMAP) barometer and the empirical calibration technique. The GMAP reaction has been empirically calibrated several times with varying results, due in large part to the choice of solution models. The method's dependence on the choice of solution model makes it an excellent was in which to address the quality of specific models. By repeating the calibration exercise for different combinations of solution models, it is possible to determine which set of solution models results in the lowest uncertainties, using a sample set of 72 published electron microprobe analyses of pelitic rocks containing the assemblage garnet+muscovite+biotite +plagioclase+aluminum silicate+quartz. As an independent test of the suitability of solution models, we compare the fit of the sample set to the kyanite-sillimanite equilibrium curve, obtaining thermobarometric estimates by simultaneous solution of the garnet-biotite exchange thermometer and GMAP calibrated from thermochemical data.