The three-day eruption at Novarupta in 1912 consisted of three discrete episodes. Episode I began with plinian dispersal of rhyolitic fallout (Layer A) and contemporaneous emplacement of rhyolitic ignimbrites and associated proximal veneers. The plinian column was sustained throughout most of the interval of ash flow generation, in spite of progressive increases in the proportions of dacitic and andesitic ejecta at the expense of rhyolite. Accordingly, plinian Layer B, which fell in unbroken continuity with purely rhyolitic Layer A, is zoned from >99% to ???15% rhyolite and accumulated synchronously with emplacement of the correspondingly zoned ash flow sequence in Mageik Creek and the Valley of Ten Thousand Smokes (VTTS). Only the andesiterichest flow units that cap the flow sequence lack a widespread fallout equivalent, indicating that ignimbrite emplacement barely outlasted the plinian phase. On near-vent ridges, the passing ash flows left proximal ignimbrite veneers that share the compositional zonation of their valley-filling equivalents but exhibit evidence for turbulent deposition and recurrent scour. Episode II began after a break of a few hours and was dominated by plinian dispersal of dacitic Layers C and D, punctuated by minor proximal intraplinian flows and surges. After another break, dacitic Layers F and G resulted from a third plinian episode (III); intercalated with these proximally are thin intraplinian ignimbrites and several andesite-rich fall/flow layers. Both CD and FG were ejected from an inner vent <400 m wide (nested within that of Episode I), into which the rhyolitic lava dome (Novarupta) was still later extruded. Two finer-grained ash layers settled from composite regional dust clouds: Layer E, which accumulated during the D-F hiatus, includes a contribution from small contemporaneous ash flows; and Layer H settled after the main eruption was over. Both are distinct layers in and near the VTTS, but distally they merge with CD and FG, respectively; they are largely dacitic but include rhyolitic shards that erupted during Episode I and were kept aloft by atmospheric turbulence. Published models yield column heights of 23-26 km for A, 22-25 km for CD, and 17-23 km for FG; and peak mass eruption rates of 0.7-1x108, 0.6-2x108, and 0.2-0.4x108 kg s-1, respectively. Fallout volumes, adjusted to reflect calculated redistribution of rhyolitic glass shards, are 8.8 km3, 4.8 km3, and 3.4 km3 for Episodes I, II, and III. Microprobe analyses of glass show that as much as 0.4 km3 of rhyolitic glass shards from eruptive Episode I fell with CDE and 1.1 km3 with FGH. Most of the rhyolitic ash in the dacitic fallout layers fell far downwind (SE of the vent); near the rhyolite-dominated ignimbrite, however, nearly all of Layers E and H are dacitic, showing that the downwind rhyolitic ash is of 'co-plinian' rather than co-ignimbrite origin. ?? 1992 Springer-Verlag.