Seismic hazard assessments for a Cascadia subduction zone earthquake are largely based on the rupture area predictions of dislocation models constrained by geodetic and geothermal data; this paper tests the consistency of the models for the 1700 great Cascadia earthquake with compiled coastal coseismic subsidence as estimated from paleoelevation studies. Coastal estimates have large uncertainties but show a consistent pattern. Greatest coseismic subsidence (∼1–2 m) occurred in northern Oregon/ southern Washington; subsidence elsewhere was ∼0–1 m. Elastic dislocation models constrained by interseismic geodetic and thermal data are used to predict the coseismic subsidence for two likely strain accumulation periods of (i) 800 and (ii) 550 yr of plate convergence and for uniform megathrust slip of 10, 20, 30, and 50 m. The former two models provide a better and equally good fit; predicted subsidence is in broad agreement with marsh estimates. Discrepancies exist, however, at the ends of the subduction zone. In the south, misfit may be due to breakup of the Gorda plate. The discrepancy in the north may be explained if the 1700 event released only part of the accumulated strain there, consistent with long-term net uplift in excess of eustatic sea-level rise. The coseismic slip magnitude, estimated by comparing uniform slip model predictions with marsh coseismic subsidence estimates, is consistent with the M 9 earthquake indicated by Japanese tsunami records. The coseismic slip was greatest in northern Oregon/southern Washington, declining to the north and south.