Here we assess estimates of atmospheric evaporative demand over China in 12 state‐of‐the‐art global climate models (GCMs) against observed D20 pan evaporation (Epan) over the period of 1961–2000. To do that, we use an energy‐relevant and physical‐based approach, namely, PenPan model, to comprehensively evaluate GCM performance with respect to their ability to simulate annual, seasonal, and monthly statistics of Epan (and its radiative and aerodynamic components, Ep,R and Ep,A). The results indicated that most GCMs generally captured the spatial pattern and seasonal cycle of Epan, Ep,R, and Ep,A. However, regional means of annual and monthly Epan, Ep,R, and Ep,A were underestimated by most GCMs mainly due to negatively biased surface air temperature (Ta) and vapor pressure deficit (vpd) outputted/simulated by the GCMs. Overall, the discrepancies among GCMs in estimating the regional statistics (regional means and seasonal cycles) of Ep,A were relatively larger than that of Ep,R, which indicates considerable uncertainties in the calculation of the aerodynamic component of evaporation based on the GCM outputs. Moreover, a few GCMs captured negative trends of regional mean annual and seasonal Epan, Ep,R, and Ep,A well over the period of 1961– 2000, but most showed positive trends. The underestimation of net radiation (Rn) and overestimation of wind speed at 2 m (u2) in most GCMs may, to some extent, accentuate/compensate the negative biases in GCM‐estimated annual and seasonal Epan, Ep,R, and Ep,A. The results demonstrate the importance of incorporating observation of pan evaporation and well‐validated PenPan model to evaluate GCM performance on atmospheric evaporative demand that is relevant to projections of future drought and regional water‐energy budgets.