Throwing motion is a complicated whole-body movement. The kinematics and dynamics of throwing have been analyzed and simulated in the literature mostly from the experimental measurements such as motion capture. In this presentation, the overarm throwing motion of a 55 degree-offreedom biped human model is generated from multibody dynamics and optimization. The whole-body dynamic balance is maintained during the motion through an efficient Zero- Moment Point formulation. The proposed method is completely predictive and no input reference motion is required. As outputs, the generated motion and the required actuator torques are calculated. The results show dynamically feasible human motion of throwing in a physics-based simulation environment. Key words: motion generation, multibody dynamics, optimization, Zero-Moment Point, throwing.