Many peraluminous igneous rocks are characterized by the presence of Al-rich minerals such as cordierite, garnet and aluminosilicate polymorphs. Clearly, one of the important thermodynamic intensive variables controlling the stability of these minerals, relative to Al-poor phases such as ortho- and clinopyroxene, must be the activity of Al 2O 3 ( a Al 2O 3 ) in the melt. Calculated phase relationships of anhydrous mineral assemblages in equilibrium with quartz-saturated liquids show that, at crustal pressures (less than about 10 kbar), garnet, cordierite, and aluminosilicate are only likely to crystallize from melts in which a Al 2O 3 is up to one order of magnitude higher than that required for orthopyroxene+clinopyroxene crystallization. In spite of this, peraluminous granitoid liquids saturated with orthopyroxene+garnet or orthopyroxene+cordierite do not ordinarily contain significantly larger amounts of excess Al (= Al in excess of that required to form feldspars) than peraluminous liquids saturated with orthopyroxene+clinopyroxene. These observations indicate that there must be profound compositional controls on the activity coefficient of Al in silicic melts, and the strongest of such controls appears to be the total alkali content of the melt. The activity coefficient of Al in silicic melts varies directly with total alkali oxide content, so that alkali-poor melts can become markedly peraluminous without ever becoming saturated in aluminous phases, whereas alkali-rich melts of similar normative corundum content can crystallize garnet, cordierite or even aluminosilicate. Because activity of alumina in a silicic melt is not a simple function of excess Al content, classification schemes for silicic igneous rocks based on model mineralogical assemblages, which reflect the value of the thermodynamic effective concentration of Al during crystallization, convey more petrologic information than those based on chemical parameters such as normative corundum content or alumina saturation index.