Compositional data for >400 pumice clasts, organized according to eruptive sequence, crystal content, and texture, provide new perspectives on eruption and pre-eruptive evolution of the >600 km³ of zoned rhyolitic magma ejected as the Bishop Tuff during formation of Long Valley caldera. Proportions and compositions of different pumice types are given for each ignimbrite package and for the intercalated plinian pumice-fall layers that erupted synchronously. Although withdrawal of the zoned magma was less systematic than previously realized, the overall sequence displays trends toward greater proportions of less evolved pumice, more crystals (0·5–24 wt %), and higher FeTi-oxide temperatures (714–818°C). No significant hiatus took place during the 6 day eruption of the Bishop Tuff, nearly all of which issued from an integrated, zoned, unitary reservoir. Shortly before eruption, however, the zoned melt-dominant portion of the chamber was invaded by batches of disparate lower-silica rhyolite magma, poorer in crystals than most of the resident magma but slightly hotter and richer in Ba, Sr, and Ti. Interaction with resident magma at the deepest levels tapped promoted growth of Ti-rich rims on quartz, Ba-rich rims on sanidine, and entrapment of near-rim melt inclusions relatively enriched in Ba and CO₂. Varied amounts of mingling, even in higher parts of the chamber, led to the dark gray and swirly crystal-poor pumices sparsely present in all ash-flow packages. As shown by FeTi-oxide geothermometry, the zoned rhyolitic chamber was hottest where crystal-richest, rendering any model of solidification fronts at the walls or roof unlikely. The main compositional gradient (75–195 ppm Rb; 0·8–2·2 ppm Ta; 71–154 ppm Zr; 0·40–1·73% FeO*) existed in the melt, prior to crystallization of the phenocryst suite observed, which included zircon as much as 100 kyr older than the eruption. The compositions of crystals, though themselves largely unzoned, generally reflect magma temperature and the bulk compositional gradient, implying both that few crystals settled or were transported far and that the observed crystals contributed little to establishing that gradient. Upward increases in aqueous gas and dissolved water, combined with the adiabatic gradient (for the ∼ 5 km depth range tapped) and the roofward decline in liquidus temperature of the zoned melt, prevented significant crystallization against the roof, consistent with dominance of crystal-poor magma early in the eruption and lack of any roof-rind fragments among the Bishop ejecta, before or after onset of caldera collapse. A model of secular incremental zoning is advanced wherein numerous batches of crystal-poor melt were released from a mush zone (many kilometers thick) that floored the accumulating rhyolitic melt-rich body. Each batch rose to its own appropriate level in the melt-buoyancy gradient, which was self-sustaining against wholesale convective re-homogenization, while the thick mush zone below buffered it against disruption by the deeper (non-rhyolitic) recharge that augmented the mush zone and thermally sustained the whole magma chamber. Crystal–melt fractionation was the dominant zoning process, but it took place not principally in the shallow melt-rich body but mostly in the pluton-scale mush zone before and during batchwise melt extraction.