A simple heat balance model for an evolving magma chamber is used to make predictions of the time scales for magma differentiation, which are compared with geological and isotopic constraints on the rates of crystallization and differentiation. In a 10 km3 magma chamber releasing thermal energy at a rate of 100 MW, basalt and rhyolite magmas should reach 50% crystallization after ~2500 and <1500 years, respectively. The tendency for phenocrysts to remain suspended in a cooling magma increases with melt viscosity and hence the degree of magma differentiation. The time scales of crystallization and differentiation may be estimated on the basis of the U-series isotope compositions of separated crystals and bulk rocks, crystal size distributions, and trace element and Sr isotope profiles in phenocryst phases. The last of these indicate crystal residence ages in the melt of tens of years up to 100 years. Short residence and even differentiation times are also obtained from Ra-Th isotope studies of alkali feldspars, and rocks that experienced alkali feldspar fractionation. However, much older ages of 103-105 years for separated phases have been reported from recent volcanic centres in St Vincent in the Lesser Antilles, Vesuvius in Italy, the Kenya Rift Valley and Long Valley, California. These old crystal ages are all from relatively evolved igneous rocks, as predicted from their higher melt viscosities and simple models of cooling and crystal settling. However, the old ages are also typically obtained for complex minerals that are not in bulk equilibrium with their host rocks, and so, apart from offering a minimum estimate of the age of the particular magmatic system, their significance for models of differentiation of the host magmas is not clear. An alternative approach is therefore to determine the variations in U-Th-Ra isotope compositions of bulk rocks reflecting different degrees of magma differentiation, and such data indicate that differentiation in more mafic magmas takes much longer than in more evolved magmas. For example, 50% fractional crystallization of basanite to produce phonolite on Tenerife took ~105 years, whereas a further 50% fractional crystallization to generate the more evolved phonolites occurred within a few hundred years of eruption. On Tenerife the more mafic magmas fractionated at greater depths, and the rates of fractional crystallization were higher in the more evolved magmas studied. This is readily explained by a cooling model in which a large volume of primitive magma deep in the crust has a longer cooling time than a smaller body of differentiated magma at shallower depths in the crust.