Evapotranspiration (ET) combines the land‐atmosphere water, energy, and carbon cycles and plays a critical role in climate studies. However, ET is difficult to quantify accurately. This study compares three 1982–2013 ET data sets for China based on surface water balance values, a modified Penman‐Monteith (MPM) model and reanalysis data. Water balance ET data, which have been widely used as benchmark data for regional ET, were calculated with and without considering reservoir water storage. MPM ET values were calculated from combinations of input data from different sources, including solar radiation and meteorological data. The MPM model is primarily energy determined, although the impacts of vegetation, soil moisture, air temperature, and wind speed on ET are also considered. The reanalysis ET data used in this study are derived from atmospheric and off‐line land reanalyses. We examined their spatial patterns, interannual variabilities, trends, and key drivers. We found that MPM ET is consistent with water balance ET in detecting ET trends for 1997 to 2013, whereas reanalysis ET is not. The trends and interannual variabilities of ET found via reanalyses are primarily determined by precipitation patterns. However, the precipitation data derived from the atmospheric reanalyses present significant uncertainties. Off‐line land reanalyses correct these biases but do not consider terrestrial water storage changes resulting from reservoir construction and interannual variability in vegetation. Both omissions may produce unreliable trends of ET in reanalyses for China. This study will help improve current ET simulations and contribute to further extending climate studies.