Although there is theoretical and empirical support for the hypothesis that some ecosystems show abrupt, nonlinear responses to perturbation, predicting when an ecosystem is approaching a critical threshold or shifting to an alternative stable state is challenging. We hypothesize that the loss of ecological resilience caused by long-term exposure to anthropogenic perturbation will decrease the size of “attraction basins”, thereby increasing the likelihood of a state shift. Disturbed communities also have a tendency to remain in an alternative stable state. Therefore, quantifying recovery thresholds, defined as the stressor level that must be achieved to restore ecological integrity, is a critical component of stream restoration. We used results of stream microcosm experiments, spatially extensive surveys, and long-term monitoring of a metal-polluted stream to identify ecological thresholds for benthic communities. We estimated thresholds using conventional piecewise regression approaches, multivariate techniques, and a new nonparametric method (SiZer, significant zero crossings). Results showed dramatic shifts in community composition when experimentally derived thresholds were exceeded; however, benthic communities quickly recovered when stressors were reduced below threshold levels. We discuss the implications of these results for chronically disturbed communities that are simultaneously experiencing effects of global climate change.