In June 1912, the world's largest twentieth century eruption broke out through flat-lying sedimentary rocks of Jurassic age near the base of Trident volcano on the Alaska Peninsula. The 60 h ash-flow and Plinian eruptive sequence excavated and subsequently backfilled with ejecta a flaring funnel-shaped vent since called Novarupta. The vent is adjacent to a cluster of late Quaternary stratocones and domes that have released about 140 km³ of magma in the past 150 k.y. Although the 1912 vent is closest to the Trident group and is also close to Mageik and Griggs volcanoes, it was the summit of Mount Katmai, 10 km east of Novarupta, that collapsed during the eruption to form a 5.5 km³ caldera. Many earthquakes, including 14 in the range M 6−7, took place during and after the eruption, releasing 250 times more seismic energy than the 1991 caldera-forming eruption of the Philippine volcano, Pinatubo. The contrast in seismic behavior may reflect the absence of older caldera faults at Mount Katmai, lack of upward (subsidence opposing) magma flow owing to lateral magma withdrawal in 1912, and the horizontally stratified structure of the thick shale-rich Mesozoic basement. The Katmai caldera compensates for only 40% of the 13 km³ of 1912 magma erupted, which included 7–8 km³ of slightly zoned high-silica rhyolite and 4.5 km³ of crystal-rich dacite that grades continuously into 1 km³ of crystal-rich andesite. We have now mapped, sampled, and studied the products of all 20 components of the Katmai volcanic cluster. Pyroxene dacite and silicic andesite predominate at all of them, and olivine andesite is also common at Griggs, Katmai, and Trident volcanoes, but basalt and rhyodacite have erupted only at Mount Katmai. Rhyolite erupted only in 1912 and is otherwise absent among Quaternary products of the cluster. Pleistocene products of Mageik and Trident and all products of Griggs are compositionally distinguishable from those of 1912 at Novarupta. Holocene products of Mount Martin and Trident are closer in composition to the andesite-dacite array of 1912, but they reveal consistent differences. The affinity of the 1912 suite is closest with the array of products erupted by the Southwest Katmai cone, the edifice that had produced the only pre-1912 rhyodacite as well as the largest prehistoric Plinian eruption in the cluster. It is doubtful that any 1912 magma had been stored beneath Novarupta or Trident, and there is no evidence that more than one magma chamber erupted in 1912. Despite a compositional gap separating the aphyric rhyolite from the very crystal-rich andesite-dacite continuum, isotopic and chemical affinities linking all the 1912 ejecta and the continuity of all those ejecta in magmatic temperature and oxygen fugacity suggest that the rhyolite originated principally by incremental upward expulsion of interstitial melt from subjacent andesite-dacite mush. A large reservoir of such hot crystal mush is required both as the residue of rhyolitic melt separation and as a proximate heat source to thermally sustain the nearly aphyric condition of the overlying rhyolite. A model is presented for a unitary zoned chamber beneath Mount Katmai.