Much of the uncertainty in determining the number and magnitude of past great earthquakes in the Cascadia subduction zone of western North America stems from difficulties in using estuarine stratigraphy to infer the size and rate of late Holocene relative sea-level changes. A sequence of interbedded peaty and muddy intertidal sediment beneath a small, protected tidal marsh in a narrow inlet of Coos Bay, Oregon, records ten rapid to instantaneous rises in relative sea level. Each rise is marked by a contact that records an upward transition from peaty to muddy sediment. But only two contacts, dating from about 1700 and 2300 yr ago, show the site-wide extent and abrupt changes in lithology and foraminiferal and diatom assemblages that can be used to infer at least half a meter of sudden coseismic subsidence. Although the characteristics of a third, gradual contact do not differ from those of some contacts produced by nonseismic processes, regional correlation with other similar sequences and high-precision 14C dating suggest that the third contact records a great plate-boundary earthquake about 300 yr ago. A fourth contact formed too slowly to have been caused by coseismic subsidence. Because lithologic and microfossil data are not sufficient to distinguish a coseismic from a nonseismic origin for the other six peat-mud contacts, we cannot determine earthquake recurrence intervals at this site. Similar uncertainties in great earthquake recurrence and magnitude prevail at similar sites elsewhere in the Cascadia subduction zone, except those with sequences showing changes in fossils indicative of >1 m of sudden subsidence, sand sheets deposited by tsunamis, or liquefaction features.