AbstractAlthough many potential evapotranspiration (PET) formulas are available, there is still a lack of knowledge on when and where to use them in catchment modeling world‐wide. Here we experimented with three different formulas in a global hydrological model (the World‐wide HYPE), using 15 years of observations from 5,338 streamflow gauges and global evapotranspiration from Earth‐observations (MOD16). We tested model performance in a multi‐process approach to select the best formula for catchments covering the global landmass. From comparing the results with land‐cover, climate classification, water‐energy limitations, we found that climate is the main driver behind the spatial patterns in model performance. Hargreaves was the best PET formula in 50% of the catchments, most of them located in the Amazonas, central Europe, and Oceania; Jensen‐Haise was better for catchments in northern latitudes (36%). Finally, Priestley‐Taylor was the best formula for India and latitudes above 65° North. The selection of a PET formula seems to be more critical in tropical regions close to the equator, where the differences in performance are above 50%. This is also where PET is highest. We found a strong connection between the five main Köppen regions and the PET formulas, further supported by landcover analysis. Hence, the PET formulas differed in their capacity to provide useful input to the water balance modeling, with complex formulas only giving improved predictions in temperate and polar regions; however, for the rest of the globe simpler formulas were better. We thus recommend to apply different PET formulas based on climatic regions world‐wide.