In the field of harvesting robot control with vision feedback, hand-eye system calibration is very important which directly affects the precision of the task. Aiming at the feature that the vision of harvesting robot is fixed, a hand-eye system calibration method was developed. In the proposed method, by making use of the stereo vision to obtain coordinates of the target point in the robot coordinate system, and combining robot motion parameters, the calibration equations can be established first, and the eye-to-hand systems can be further calibrated based on improved least square method. This calibration process is simple, and does not need complex auxiliary devices. Related experiments have been done based on the suggested method to obtain the transformation matrix between hand-eye systems calibration, which has been shown to be of precision and efficiency.