A new method aiming at reducing peak torques is developed in this paper for kinematically redundant manipulators. The method considers the contributions of both velocity and acceleration to the joint torque. Instead of minimizing the joint torque or kinetic energy at the current station, the method always uses the current torque to approach an optimum velocity of the following station, which, along with joint acceleration, minimizes the torque of the following station, while at the same time keeps the current torque within the limits. Simulations are conducted for three cases and the results show the clear superiority of the proposed method over the conventional torque optimization and energy minimization methods.