The width of plate-boundary fault rupture at the Cascadia subduction zone, a dimension related to earthquake magnitude, remains uncertain because of the lack of quantitative information about land-level movements during past great-earthquake deformation cycles. Beneath a marsh at Alsea Bay, on the central Oregon coast, four sheets of tsunami-deposited sand blanket contacts between tidal mud and peat. Radiocarbon ages for the sheets match ages for similar evidence of regional coseismic subsidence and tsunamis during four of Cascadia's great earthquakes. Barring rapid, unrecorded postseismic uplift, reconstruction of changes in land level from core samples using diatom and foraminiferal transfer functions includes modest coseismic subsidence (0.4??0.2 m) during the four earthquakes. Interpretation is complicated, however, by the 30-38% of potentially unreliable transfer function values from samples with poor analogs in modern diatom and foraminiferal assemblages. Reconstructions of coseismic subsidence using good-analog samples range from 0.46??0.12 to 0.09??0.20 m showing greater variability than implied by sample-specific errors. From apparent high rates of land uplift following subsidence and tsunamis, we infer that postseismic rebound caused by slip on deep parts of the plate boundary and (or) viscoelastic stress relaxation in the upper plate may be almost as large as coseismic subsidence. Modest coseismic subsidence 100 km landward of the deformation front implies that plate-boundary ruptures in central Oregon were largely offshore. Ruptures may have been long and narrow during earthquakes near magnitude 9, as suggested for the AD 1700 earthquake, or of smaller and more variable dimensions and magnitudes. ?? 2008 Elsevier Ltd. All rights reserved.