The Yellowstone Plateau volcanic field has undergone repeated eruption of rhyolitic magma strongly depleted in 18O. Large calderas subsided 2.0, 1.3, and 0.6 Ma ago, on eruption of ash flow sheets that represent at least 2500, 280, and 1000 km3 of zoned magma. More than 60 other rhyolite lavas and tuffs permit reconstruction of the long-term chemical and isotopic evolution of the silicic system. Narrow Δ18O ranges in the ash flow sheets contrast with wide Δ18O variations in postcaldera lavas of the first and third caldera cycles. Earliest postcollapse lavas are 3 to 6% lighter than the proceding ash flow sheets. The 18O depletions were short-lived events that immediately followed caldera subsidence; hundreds of cubic kilometers of magma were drastically 18O depleted and thousands were depleted by 1--2%. Sequences of postcaldera lavas record partial recovery toward precaldera Δ18O values; secular trends between collapse events thus reflect gradual reenrichment of the roofmost magma in 18O. Much of the subcaldera reservoir was affected, because lavas that erupted as far apart as 115 km reflect the same pattern of depletion and partial recovery. Contemporaneous extracaldera rhyolites have the highest Δ18O values in the volcanic field and show no effects of the repeated depletions. Sr and Pb isotope ratios of intracaldera rhyolites jump to more radiogenic values at times of caldera formaiton and show a longterm zigzag pattern like that of Δ18O. Although some contamination by foundering roof rocks seems to be required, water was probably the predominant contaminant. Even if roof rocks had been strongly depleted in 18O before engulfment, their assimilation would have been far from sufficient to account for the large 18O shift. The low- 18O lavas contain no xenocrysts and show no trace element or phenocryst evidence of massive contamination. Their Fe-Ti-oxide temperatures indicate no cooling relative to the caldera-forming ash flow magma, and their whole-rock, glass, and phenocryst chemistry suggests compositional continuity, with the ash flow sequence. Oxygen exchange between the magma and a mass of low- 18O water greatly exceeding solubility limits may require (1) recurrent explosive activity to sustain access and mixing of water with the magma and (2) convection of the magma reservoir to prevent local saturation.