Objective:To determine the extent of inter-observer variation in delineation of heart after and before studying the heart atlas and its impact on dose distribution of the whole heart and every cardiac valve in radical radiotherapy of patients with esophagus cancer in blind modes. Methods:Ten patients with esophagus cancer and without operation were asked to contour the heart in computed tomography transverse section in blind modes by five observers. Inter-observer variability were compared in the delineation of the whole heart and estimated doses for heart and every cardiac valve before and after studying the heart atlas. Results:Compared with the mean heart volume without guidelines,the mean heart volume with guidelines was increased from 635.3 cm3 to 647.6 cm3. Soensen-Dice similarity index(DSI)was increased from 0.84~0.87 to 0.92~0.94. Jaccard similarity index(JSI)was increased from 0.91~0.93 to 0.96~0.97. After learning guidelines,the volume of the heart was increased significantly,I_Tot/U_ Tot increased from 0.75 ± 0.02 to 0.84 ± 0.03. In addition to the heart and the pulmonary valve,Dmean and Dmax of aortic valve,mitral valve,tricuspid valve were increased significantly after learning guidelines. The coefficients of variation(CV)of Dmean and Dmax of hart and valves were increased significantly. Conclusion:There was a significant difference in the delineation of heart for different observers,which could be decreased by the introduction of heart atlas. The delineating inter-observer variability has no dosimetric effects on the whole heart and pulmonic valve,leads to the difference of Dmean and Dmax of the estimated dose on aortic valve,mitral valve,tricuspid valve. Furthermore,the introduction of heart atlas decreases the difference of radiation dose among the above structures.