Hydrologic models that rely on site specific linear and non-linear regression water temperature (Tw) subroutines forced solely with observed air temperature (Ta) may not accurately estimate Tw in mixed-use urbanizing watersheds where hydrogeological and land use complexity may confound common Tw regime assumptions. A nested-scale experimental watershed study design was used to test Tw model predictions in a representative mixed-use urbanizing watershed of the central USA. The linear regression Tw model used in the Soil and Water Assessment Tool (SWAT), a non-linear regression Tw model, and a process-based Tw model that accounts for watershed hydrology were evaluated. The non-linear regression Tw model tested at a daily time step performed significantly (P < 0.01) better than the linear Tw model currently used in SWAT. Both regression Tw models overestimated Tw in lower temperature ranges (Tw < 10.0 °C) with percent bias (PBIAS) values ranging from −28.2% (non-linear Tw model) to −66.1% (linear regression Tw model) and underestimated Tw in the higher temperature range (Tw > 25.0 °C) by 3.2%, and 7.2%, respectively. Conversely, the process-based Tw model closely estimated Tw in lower temperature ranges (PBIAS = 4.5%) and only slightly underestimated Tw in the higher temperature range (PBIAS = 1.7%). Findings illustrate the benefit of integrating process-based Tw models with hydrologic models to improve model transferability and Tw predictive confidence in urban mixed-land use watersheds. The findings in this work are distinct geographically and in terms of mixed-land use complexity and are therefore of immediate value to land-use managers in similarly urbanizing watersheds globally. Copyright © 2015 John Wiley & Sons, Ltd.