Stream temperature models are commonly evaluated using conventional metrics such as root mean square error and mean bias error, which provide a useful but incomplete assessment of model performance. While these metrics quantify overall agreement between observed and simulated temperature, they may not adequately evaluate the model’s ability to simulate various aspects of the thermal regime, such as seasonal extremes, timing, duration, and frequency of warm events. Rigorously modelling these components of the stream thermal regime can be important since these temperature dynamics influence aquatic habitat suitability and biogeochemical functioning. Using thermal signatures to evaluate models is one approach to overcome limitations of more conventional evaluation metrics. While the use of signature indices has become widespread for hydrological model evaluation, a similar framework has not yet been developed for thermal model evaluation. In this study, we developed and applied a signature-based framework for stream temperature model evaluation. The framework includes signature indices related to seasonal dynamics, warm event dynamics, and energy budget partitioning. Stream temperature was modelled using the thermal module of the Raven hydrological modelling framework for the Ouelle River (844 km², Quebec, Canada), which has been intensively monitored at multiple sites for 15 years. The proposed signature indices were evaluated alongside conventional metrics to provide a comprehensive and process-oriented assessment of model performance. Overall, this refined evaluation approach improves stream temperature model evaluation by more effectively capturing physical realism, thereby increasing the robustness of thermal model applications under changing environmental conditions.