Low summer river flows can increase vulnerability to warming, impacting coldwater fish. Water managers need tools to quantify the complex linkages between flow and water temperature, yet statistical models often assume a constant relationship between these variables. In California’s snowmelt and groundwater-influenced Scott River where agricultural irrigation consumes most summer river flow, flow variation had stronger effects on water temperature in April–July than other months. Using 24 years of daily air temperature and flow data as predictors, we compared multiple statistical methods for modeling daily Scott River water temperatures, including generalized additive models with non-linear interactions between flow and day of the year. Models with seasonally varying flow effects performed better than those assuming a constant relationship between water temperature and flow. Cross-validation root mean squared errors of the selected models were ≤1 °C. We applied the models to several instream flow scenarios currently being considered by stakeholders and regulatory agencies. Relative to historic conditions, the most protective flow scenario would reduce average annual maximum temperature from 25.9 °C to 24.6 °C, reduce average annual degree-days exceedance of 22 °C (a cumulative thermal stress metric) from 107 to 54, and delay the onset of water temperatures greater than 22 °C during some drought years. Withdrawal of river water after 1 June, including for groundwater management purposes, could contribute to additional exceedances of 22 °C. These methods can be applied to model any stream with long-term flow and water temperature measurements, with applications including scenario prediction and infilling data gaps.