Urbanization in watersheds typically increases the frequency and intensity of disruptive high-flow events in streams. Stormwater managers take actions to reduce peak flows and restore stream functioning, and the outcomes of those actions depend on how metrics of ecosystem structure and function respond to peak flow. Our goal was to understand subsidy-stress responses by whole-stream ecosystem metabolism to high-flow events of different magnitudes. Specifically, we aimed to determine thresholds of stream discharge at which ecosystem functions transition from resisting to being disturbed by high flow. We used 3 y of high-frequency dissolved O2 (DO), stream discharge, and solar irradiance data in 7 urban streams near Cleveland, Ohio, and Denver, Colorado, USA. We estimated daily rates of gross primary production (GPP), ecosystem respiration (ER), and gas exchange velocity through inverse Bayesian modeling of DO dynamics (2695 stream days modeled). We delineated >900 high-flow events from baseflow conditions, and for 228 isolated events (i.e., no prior event within 72 h), we calculated maximum discharge to predict changes in GPP and ER. We also placed tracer particles and tiles into Mill Creek, Ohio, to test whether observed disturbance in GPP was related to scour of primary producer biomass. In all streams, GPP was more sensitive than ER to high flow, and disturbance thresholds varied widely (0.1-86 m3/s), with lower thresholds in semiarid Denver and in one Cleveland stream with a bedrock-dominated stream bed. Event-caused subsidies (i.e., increases in metabolic rate) were more common for ER than GPP. Mobilization of tracer particles and removal of biomass from tiles corresponded well with changes in GPP, suggesting that scour was the primary mechanism for stress responses. Overall, stream metabolism may serve as a useful metric for monitoring biological condition in urban streams and setting targets for remediation efforts aimed at restoring ecosystem function.