The study of the visual behaviors of observers in subjective image quality assessment is helpful in understanding of human visual system. It can also be used to improve the reliability of assessment results. In this paper, we propose a subjective image quality assessment system based on eye tracker. The system can record the viewpoints of observers' eye movements in the process of experiment. The system also can analyze the visual behaviors of observers. In the experiment, we first verify the accuracy of the recorded viewpoints. After, we choose the observers with different professional backgrounds. From the analysis of the assessment results and the corresponding viewpoints, we can see that non-experts are more easily to be attracted by the image contents, and the viewpoints of experts are more overall.