Existing comparative studies on conceptual hydrological model structures have generally focused on streamflow, with less attention to evapotranspiration (ET). Studies have shown that an appropriate ET representation in conceptual models is crucial for obtaining reliable hydrological simulations and predictions, especially under climate change and for ungauged basins. To address this gap, we investigated 30 conceptual models in terms of how they represent ET process and their ability on reproducing two state-of-the-art ET products (PML-V2 and FLUXCOM-X-BASE) in 507 catchments across different climates and landscapes. We found significant differences in ET simulation performance between models with different ET representations. Models employing a linear-relationship between ET from soil and soil moisture storage generally outperformed those that equate ET from soil directly with potential ET (PET). For each catchment, at least one model could well reproduce at least one ET product. Models’ performance in the ET simulation varied with climatic and vegetation characteristics. In humid catchments with summer-dominated rainfall or mild rainfall seasonality, a concise linear-type ET representation could effectively replicate ET products. Adding parameter(s) controlling the long-term ratio between ET and PET could further improve performance. In humid winter-rainfall-dominated catchments, explicitly representing an interception evaporation (Ei) component and constraining Ei by the interception capacity or storage was critical for reproducing ET products. In some arid catchments, considering the contribution of lower storage to ET generation was necessary to reproducing ET products. This study provides insights into how to appropriately represent the conversion process from PET and precipitation to ET in various climates.