The role of tree mortality in the global carbon balance is complicated by strong spatial and temporal heterogeneity that arises from the stochastic nature of carbon loss through disturbance. Characterizing spatio-temporal variation in mortality (including disturbance) and its effects on forest and carbon dynamics is thus essential to understanding the current global forest carbon sink, and to predicting how it will change in future. We analyzed forest inventory data from the eastern United States to estimate plot-level variation in mortality (relative to a long-term background rate for individual trees) for nine distinct forest regions. Disturbances that produced at least a four-fold increase in tree mortality over an approximately five-year interval were observed in 1-5% of plots in each forest region. The frequency of disturbance was lowest in the northeast, and increased southwards along the Atlantic and Gulf coasts as fire and hurricane disturbances became progressively more common. Across the central and northern parts of the region, natural disturbances appeared to reflect a diffuse combination of wind, insects, disease, and ice storms. By linking estimated covariation in tree growth and mortality over time with a data-constrained forest dynamics model, we simulated the implications of stochastic variation in mortality for long-term aboveground biomass changes across the eastern US. A geographic gradient in disturbance frequency induced notable differences in biomass dynamics between the least- and most-disturbed regions, with variation in mortality causing the latter to undergo considerably stronger fluctuations in aboveground stand biomass over time. Moreover, regional simulations showed that a given long-term increase in mean mortality rates would support greater aboveground biomass when expressed through disturbance effects compared to background mortality, particularly for early-successional species. The effects of increased tree mortality on carbon stocks and forest composition may thus depend partly on whether future mortality increases are chronic or episodic in nature.