Abstract: It is very important for studying surface energy and water balance to improve the accuracy of evapotranspiration (ET) estimation. Based on eddy covariance measurements and microclimate observational data available, comparisons were done in accuracy of simulating ET with the FAO PenmanMonteith  model from the marshland, rice paddy and soybean field in the Sanjiang Plain. The results showed that the values of ET simulated with the model over marshland was significantly higher than the measured one (averagely 81.8% higher) when the crop coefficients recommend by FAO were adopted, and its modeling efficiency  was negative, which indicated that the ET from the marshland couldn’t be simulated by the model. While the seasonal variation of ET over rice paddy and soybean field could be simulated by the model and the accuracy in simulating ET from rice paddy was better than that from soybean field. Crop coefficients  (K_c) of marshland, rice paddy and soybean field were all significantly positively related to leaf area index, and crop coefficient of soybean field was also significantly negatively related to vapor pressure deficit. With K_c modified through linear regression, the FAO Penman-Monteith model markedly improved the estimation accuracy for marshland, rice paddy and soybean field, with the mean bias error  ranging from -0.1 to 0.3 mm·d^-1, root mean square error  ranging from 0.50 to 0.67 mm·d^-1 and modeling efficiency  ranging from 0.69 to 0.85. Still, the accuracy in simulating ET from rice paddy was superior to that from the other two underlying surfaces. The FAO PenmanMonteith model was suitable to simulate the ET from rice paddy whether the crop coefficient was modified or not. However, the crop coefficient must be modified if the model was used to simulate the ET from marshland and soybean field.