Abstract: Based on eddy covariance measurements and microclimate observations available from 2005 to 2007, the simulating accuracy of evapotranspiration with P-T model of rain-fed soybean field from May to October in Sanjiang Plain is analyzed. Results indicate that simulated values of evapotranspiration by P-T model with conventional parameter (1.26) are significantly higher than observations before emergence and during the growing season of soybean, and the mean bias error (MBE) are 1.65 mm·d^-1 and 1.22 mm·d^-1. However, simulated values are significantly lower than measurements after harvest, with the MBE of-0.74 mm·d^-1. Modeling efficiency (ME) of P-T model are all negative values, which indicates that the model cannot be used in predicting evapotranspiration of soybean field during different periods. The cause may have much to do with the parameter, which is assumed as constant value of 1.26. According to measurements of evapotranspiration, the parameter is derived and shows obviously increasing trend during the whole observation periods. Average values of parameter before emergence, during the growing season, and after harvest are 0.76, 0.86 and 2.20, respectively. It is obvious that the parameter varies according to the growing phase, and it is necessary to modify the parameter based on the measured evapotranspiration of rain-fed soybean field in Sanjiang Plain.Statistical analysis shows that leaf area index (LAI) is an important factor affecting evapotranspiration of soybean field. During the growing season, the parameter is creased with increasing LAI, following a logarithmic equation and a positive correlation. Vapor pressure deficit (VPD) is the direct driving force of transporting vapor from the surface to the surrounding atmosphere. The relationship between and VPD can be described empirically by a piecewise function: When the VPD is greater than 5.05 hPa, it's a positive power function, but when the VPD is lower than 5.05, it's a negative power function. The parameter is positively related to solar radiation and negatively related to VPD before soybean emergency and is positively related to wind speed after soybean harvest.With parameter modified by using linear or non-linear regression equation, the estimation accuracy of P-T model under different periods are improved markedly. Before soybean emergency, MBE and root mean square error (RMSE) are 0.06 mm·d^-1 and 0.60 mm·d^-1, reduced by 96.4% and 71.4%, respectively. ME is improved from a negative to a positive value (0.57), close to the ideal value of 1. During the growing season, MBE and RMSE are 0.15 mm·d^-1 and 0.92 mm·d^-1, reduced by 87.7% and 38.3%, respectively, and ME from a negative to a positive value (0.28). After soybean harvest, MBE and RMSE are-0.21 mm·d^-1 and 0.41 mm·d^-1, reduced by 71.6% and 52.3%, respectively, ME turns from a negative into a positive value (0.42). It indicates that the modified P-T model can simulate the evapotranspiration of soybean field. In conclusion, P-T model is suitable to simulate the evapotranspiration only when the parameter is modified.