Forest mortality constitutes a major uncertainty in projections of climate impacts on terrestrial ecosystems and carbon-cycle feedbacks. Recent drought-induced, widespread forest die-offs highlight that climate change could accelerate forest mortality with its diverse and potentially severe consequences for the global carbon cycle, ecosystem services, and biodiversity. How trees die during drought over multiple years remains largely unknown and precludes mechanistic modeling and prediction of forest die-off with climate change. Here, we examine the physiological basis of a recent multiyear widespread die-off of trembling aspen (Populus tremuloides) across much of western North America. Using observations from both native trees while they are dying and a rainfall exclusion experiment on mature trees, we measure hydraulic performance over multiple seasons and years and assess pathways of accumulated hydraulic damage. We test whether accumulated hydraulic damage can predict the probability of tree survival over 2 years. We find that hydraulic damage persisted and increased in dying trees over multiple years and exhibited few signs of repair. This accumulated hydraulic deterioration is largely mediated by increased vulnerability to cavitation, a process known as cavitation fatigue. Furthermore, this hydraulic damage predicts the probability of interyear stem mortality. Contrary to the expectation that surviving trees have weathered severe drought, the hydraulic deterioration demonstrated here reveals that surviving regions of these forests are actually more vulnerable to future droughts due to accumulated xylem damage. As the most widespread tree species in North America, increasing vulnerability to drought in these forests has important ramifications for ecosystem stability, biodiversity, and ecosystem carbon balance. Our results provide a foundation for incorporating accumulated drought impacts into climate–vegetation models. Finally, our findings highlight the critical role of drought stress accumulation and repair of stress-induced damage for avoiding plant mortality, presenting a dynamic and contingent framework for drought impacts on forest ecosystems.