Understanding the relationship between environmental factors and vital rates is an important step in predicting a species’ response to environmental change. Species associated with sea ice are of particular concern because sea ice is projected to decrease rapidly in polar environments with continued levels of greenhouse gas emissions. The relationship between sea ice and the vital rates of the Spectacled Eider, a threatened species that breeds in Alaska and Russia and winters in the Bering Sea, appears to be complex. While severe ice can impede foraging for benthic prey, ice also suppresses wave action and provides a platform on which eiders roost, thereby reducing thermoregulation costs. We analyzed a 23‐year mark‐recapture dataset for Spectacled Eiders nesting on Kigigak Island in western Alaska, and tested survival models containing different ice and weather‐related covariates. We found that much of the variation in eider survival could be explained by the number of days per year with >95% sea ice concentration at the Bering Sea core wintering area. Furthermore, the data supported a quadratic relationship with sea ice rather than a linear one, indicating that intermediate sea ice concentrations were optimal for survival. We then used matrix population models to project population trajectories using General Circulation Model (GCM) outputs of daily sea ice cover. GCMs projected reduced sea ice at the wintering area by year 2100 under a moderated emissions scenario (RCP 4.5) and nearly ice‐free conditions under an unabated emissions scenario (RCP 8.5). Under RCP 4.5, stochastic models projected an increase in population size until 2069 coincident with moderate ice conditions, followed by a decline in population size as ice conditions shifted from intermediate to mostly ice‐free. Under RCP 8.5, eider abundance increased until 2040 and then decreased to near extirpation toward the end of the century as the Bering Sea became ice‐free. Considerable uncertainty around parameter estimates for survival in years with minimal sea ice contributed to variation in stochastic projections of future population size, and this uncertainty could be reduced with additional survival data from low‐ice winters.