Evapotranspiration (ET) is an important component of the surface water cycle. In this study, we analyzed the differences between the estimated ET using the crop coefficients (k(c)) recommended by the Food and Agriculture Organization (FAO) and the measured ET with a lysimeter in Loess Plateau spring wheat (Triticum aestivum L.) from 2010 to 2012. Results showed that ET estimated from FAO-recommended crop coefficients (k(c-FAO)) was significantly overestimated (55%) compared with measured ET by lysimeter. Evapotranspiration was sensitive to the relative available water (R-aw), and k(c) were strongly dependent on drought stress in semiarid area (p < 0.001). Relative available water dependence of k(c) increased with the decrease in R-aw. An improved k(c) model based on k(c-FAO )integrating drought stress factor was developed, then an empirical ET model derived from the k(c) approach was established based on the Penman-Monteith equation and the improved k model.The ET data measured by lysimeter from June of 2004 to May of 2005 were used to validate the proposed ET model. The estimated ET showed satisfactory consistency to measured ET with a linear regression coefficient of 1.01. Evapotranspiration modeling performance derived from the improved k(c) model integrating drought stress factor was superior to that derived by k(c-FAO) directly in semiarid area.