Evaporation (E) and Transpiration (T) play a critical role in water and energy budgets at regional to global scales. T is regulated via stomatal conductance (gsc), which depends on a multitude of plant physiological processes and hydrometeorological forcings. In recent years, significant advances have been made toward estimating gsc using a variety of models, ranging from relatively simple empirical models to more complex and data-intensive plant hydraulics-based models. However, a detailed assessment of the ability of these models for predicting evapotranspiration components (E and T) remains lacking. Using machine learning and eddy covariance flux tower data of 642 years, distributed across 84-sites and ten land covers globally, here we show that structural constraints in both empirical and plant hydraulics-based models of gsc limit their effectiveness for predicting evapotranspiration (ET) and its components, i.e., E and T. Notably, even when the current generation gsc models are calibrated locally, their limiting structures don’t allow them to use the information contained in the data optimally. Performance of empirical models, which are still widely used for ET estimation, is observed to be especially underwhelming for partitioning T from ET. While the plant hydraulics-based model structure is relatively effective because of its ability to capture the inextricably-linked stomatal response to soil moisture (SM) and vapor pressure deficit (VPD), we show that there still is a significant room for improvement in the structure of these models. These results underscore the need to prioritize improvements in models of gsc to constraint estimates of E and T, and thus to reduce uncertainties in assessments of plants’ role in regulating the earth’s climate.