Adopting appropriate postures during manual material-handling tasks is the key to reducing human joint injuries. Although much experimentation has been conducted in an effort to model lifting, such an approach is not general enough to consider all potential scenarios in material handling. Thus, in this paper an optimization-based motion prediction method is used to simulate realistic lifting postures and predict joint torques to evaluate the risk level of injury. A kinematically realistic digital human model has been developed such that the complicated musculoskeletal human structure is modeled as a combination of serial chains using the generalized coordinates. Lagrange's equations of motion and metabolic energy rate are derived for the digital human. The proposed method has been implemented to predict and evaluate the lifting postures based on the metabolic rate and joint torques. Our results show that different amount of external loads and tasks lead to different human postures and joint torque distribution, thus different risk level of injury.