In traditional orthodontic treatment, archwire preparation is performed by orthodontists manually. This manual pattern not only requires long time trainning of the clinicians, but also can not ensure the clinical requirement of customization and accuracy. It also increases the number of patient’s office visits and long chairside time. Thus, an archwire bending robotic system aiming to perform this task automatically was developed. The springback effect caused by high elasticity of material affects the accuracy of bend forming. In this paper, a compensation method based on an overbend prediction model and an online force detection was presented to eliminate the springback effect of archwire. This method contains two steps. Firstly, an overbend prediction model was constructed to predict the overbend allowance of a target angle, based on calculating the difference of two angles measured by a self-designed instrument and taking the angle after springback as the target angle. Secondly, a force-based bending control algorithm was designed to detect the defined zero-force state using a force sensor to online measure the force at a given target angle. Finally, bending experiments were conducted in the self-developed orthodontic archwire bending system. Results show that the proposed compensation method can eliminate springback effect to minimum and accomplish the preparation of clinical required orthodontic archwire.