The lack of river water temperature records has always been a problem in riverine ecosystem research. Where data is not available or complete, there have been many different approaches to model water temperature. While popular datasets that were constructed by deterministic models trained on multiple rivers across the globe can provide spatially extrapolated temperature estimates in data-limited regions, their performance needs to be validated prior to usage. In contrast, simple statistical models tailored to specific rivers, which rely on air-water temperature relationships, may be a simpler solution. We analyzed whether simple statistical models fit to one site can perform as well or even better than a deterministic model fit to multiple sites when predicting water temperature in the Great Lakes tributaries. Using temperature records from 10 different tributary locations across the Great Lakes watershed, we demonstrated that simple statistical models outperform the deterministic model at all sites when estimating water temperature during the growing season. A nonlinear regression using mean air temperature from the past week provides the most accurate water temperature predictions with a mean RMSE of 1.41°C and a bias close to zero.