Within the 2.8 Ma Chegem ash-flow caldera (11 ?? 15 km), a single cooling unit of rhyolitic to dacitic welded tuff more than 2 km thick is exposed in deep valleys incised during recent rapid uplift of the Caucasus Mountains. The intracaldera tuff is mineralogically fresh and unaltered, and is overlain by andesite lavas and cut by a resurgent granodiorite intrusion. Major- and trace-element compositions for a 1405-m stratigraphic section of intracaldera tuff display trends of upwardly increasing Na2O, CaO, Al2O3, total Fe, MgO, TiO2, Sr and Zr and decreasing SiO2, K2O and Rb. This mafic-upward zoning (from 76.1 to 69.9% SiO2) reflects an inverted view of the upper part of the source magma chamber. Oxygen isotope studies of 35 samples from this 1405-m section define a striking profile with "normal" igneous ??18O values (+7.0 to +8.5) in the lower 600 m of tuff, much lower ??18O values (-4.0 to +4.3) in a 700-m zone above that and a shift to high ??18O values (+4.4 to -10.9) in the upper 100 m of caldera-fill exposure. Data from two other partial stratigraphic sections indicate that these oxygen isotope systematics are probably a caldera-wide phenomenon. Quartz and feldspar phenocrysts everywhere have "normal" igneous ??18O values of about +8.5 and +7.5, respectively, whereas groundmass and glass ??18O values range from -7.7 to +12.3. Consequently, the ??18O values of coexisting feldspar, groundmass and glass form a steep array in a plot of ??feldspar vs. ??groundmass/glass. Such pronounced disequilibrium between coexisting feldspar and groundmass or glass has never before been observed on this scale. It requires a hydrothermal event involving large amounts of low-18O H2O at sufficiently high temperatures and short enough time (tens of years or less) that glass exchanges thoroughly but feldspar does not. The most likely process responsible for the O depletions at Chegem is a very high temperature (500-600??C), short-lived, vigorous meteoric-hydrothermal event that was focused within the upper 750 m of intracaldera tuff. Mass balance calculations indicate fluid fluxes of = 6 ?? 10-6 mol cm-2 s-1. We believe that the closest historical analogue to this Chegem hydrothermal event is the situation observed in the Valley of Ten Thousand Smokes (Alaska, USA), where hundreds of steam fumaroles with measured temperatures as high as 645??C persisted for 10 to 15 years in the much smaller welded ash-flow tuff sheet (??? 200 m thick) produced by the 1912 Katmai eruption.