Abstract: Based on the eddy covariance measurements and microclimate observational data available, comparison was done between the Penman, FAO Penman-Monteith (PM), Priestley-Taylor (PT) models in accuracy of simulating evapotranspiration of soybean field in the Sanjiang Plain. Results show that the values of evapotranspiration (ET) simulated with the models were all higher than the measured one when constants were adopted as parameters in modeling. The Penman model was in particular, being on average 174.6% higher. And the Penman and PM models both fell below zero in modellingmodeling efficiency (EF), which indicated that the two models were too poor in estimation accuracy to be used to predict actual ET of soybean field. In the Penman and FAO PM models, crop coefficient (K_c) was positively related to leaf area index (LAI), but negatively to saturated vapourvapor pressure difference (VPD), both at an extremely significant level; and in the PT model, α value was also positively related to LAI and wind speed and negatively to VPD, both at an extremely significant level, too. With K_c and α modified with the multiple linear regression equation, the three models improved estimation accuracy markedly, with mean bias error(MBE) ranging from -0.10 to 0.0 mm•d^-1, and root mean square errors (RMSE) and EF being 0.67 mm•d^-1 and 0.61, respectively. ANOVA further indicated that no significant difference existed between the three models in simulated value, but relatively speaking, the PT model was slightly superior to the other two in intercept, slope and MBE. Therefore, the modified PT model is an optimal one among the three for use to estimating ET of soybean field in the Sanjiang Plain.