Rhyolites can be highly evolved with Sr contents as low as 0.1 ppm and Rb Sr > 2,000. In contrast, granite batholiths are commonly comprised of rocks with Rb Sr < 10 and only rarely > 100. Mass-balance modelling of source compositions, differentiation and contamination using the trace-element geochemistry of granites are therefore commonly in error because of the failure to account for evolved differentiates that may have been erupted from the system. Rhyolitic magmas with very low Sr concentrations (???1 ppm) cannot be explained by any partial melting models involving typical crustal source compositions. The only plausible mechanism for the production of such rhyolites is Rayleigh fractional crystallization involving substantial volumes of cumulates. A variety of methods for modelling the differentiation of magmas with extremely high Rb/Sr is discussed. In each case it is concluded that the bulk partition coefficients for Sr have to be large. In the simplest models, the bulk DSr of the most evolved types is modelled as > 50. Evidence from phenocryst/glass/whole-rock concentrations supports high Sr partition coefficients in feldspars from high silica rhyolites. However, the low modal abundance of plagioclase commonly observed in such rocks is difficult to reconcile with such simple fractionation models of the observed trace-element trends. In certain cases, this may be because the apparent trace-element trend defined by the suite of cognetic rhyolites is the product of different batches of magma with separate differentiation histories accumulating in the magma chamber roof zone. ?? 1991.