Low streamflows can increase vulnerability to warming, impacting coldwater fish. Water managers need tools to quantify these impacts and predict future water temperatures. Contrary to most statistical models’ assumptions, many seasonally changing factors (e.g., water sources and solar radiation) cause relationships between flow and water temperature to vary throughout the year. Using 21 years of air temperature and flow data, we modeled daily water temperatures in California’s snowmelt-driven Scott River where agricultural diversions consume most summer surface flows. We used generalized additive models to test time-varying and nonlinear effects of flow on water temperatures. Models that represented seasonally varying flow effects with intermediate complexity outperformed simpler models assuming constant relationships between water temperature and flow. Cross-validation error of the selected model was ≤1.2 °C. Flow variation had stronger effects on water temperatures in April–July than in other months. We applied the model to predict effects of instream flow scenarios proposed by regulatory agencies. Relative to historic conditions, the higher instream flow scenario would reduce annual maximum temperature from 25.2 °C to 24.1 °C, reduce annual exceedances of 22 °C (a cumulative thermal stress metric) from 106 to 51 degree-days, and delay onset of water temperatures >22 °C during some drought years. Testing the same modeling approach at nine additional sites showed similar accuracy and flow effects. These methods can be applied to streams with long-term flow and water temperature records to fill data gaps, identify periods of flow influence, and predict temperatures under flow management scenarios.