In assessing risks related to working posture, pictures of postures are taken from various directions, which can be a source of observation error. Joint postures of the neck, lower back, knee, shoulder, and elbow were taken from 7 different viewing angles and 19 observers estimated joint angles by observing the pictures in 2-dimensional display. The joint angles were also measured using an optoelectronic motion measurement system. The estimation error increased as the viewing angle varies from the right side of the human body, but the patterns differ according to which joint angles were being observed. Guidelines to increase the validity of observation of joint angles were presented based on the results. In general, it is recommended to maintain the viewing angle within 20 degrees from the right side of the human body, while different ranges of viewing angle are recommended for each joint angle.