Long-term studies of stream ecosystems are essential for assessing restoration success because they allow researchers to quantify recovery trajectories, gauge the relative influence of episodic events, and determine the time required to achieve clean-up objectives. To quantify responses of benthic macroinvertebrate assemblages to stream remediation, we integrated results of 4 long-term (20–29 y) assessments of mining-impacted watersheds that were broadly distributed across the western US (California, Colorado, Idaho, Montana). Using a before–after control–impact (BACI) study design, we observed substantial reductions in metal concentrations and corresponding improvements of benthic assemblages following remediation. Recovery rates were relatively consistent, and streams typically recovered within 10 to 15 y after remediation was initiated (mean = 10.25 y), although episodic events changed trajectories at some sites. Differences in recovery among watersheds were likely determined by a number of factors, including the severity of contamination, effectiveness of remediation, proximity to upstream sources of colonization, and hydrologic variation. We also observed considerable variation in the rate and extent of recovery among assemblage metrics. For example, total abundance and richness recovered rapidly at most sites, but the composition of benthic macroinvertebrate assemblages remained substantially altered compared with reference sites. Using piecewise linear regression, we estimated a threshold response of Ephemeroptera, Plecoptera, and Trichoptera (EPT) species richness at ~1 cumulative criteria unit (CCU), which is the sum of the fractions of chronic water-quality criteria for metals measured, suggesting this value was protective of benthic assemblages. However, EPT richness was reduced by ~20% at 2× this CCU value, indicating that moderate exceedances of water-quality criteria could substantially affect stream biodiversity. Non-metric multidimensional scaling analyses identified common sets of species trait states across the 4 watersheds that were associated with either metal contamination or with recovering and intact reference stream assemblages. Our study illustrates the importance of long-term studies for quantifying responses to stream restoration and the usefulness of BACI designs for demonstrating cause-and-effect relationships between restoration treatments and community recovery. Because these 4 watersheds were among the most severely polluted sites in the western US, our study demonstrates the value of these investments in watershed restoration and the potential for success under the most extreme conditions.