In this dissertation, the design and implementation of a control system for stable walking of a biped robot is presented. The biped robot used as a test bed for walking experiments is called Archie that was designed and constructed in Vienna University of Technology. We also improved the robot hardware for walking by adding a joint to each ankle. The improved biped robot has 12 degrees of freedom totally, i.e. each leg has 6 joints. The proposed control system is based on decentralized control method. In this strategy, each joint's rotation angle is controlled independently and the dynamic effects of manipulator links to each other are considered as disturbances. Therefore the independent joint controller is designed such that not only the output tracks the reference trajectory but also reject the disturbance. Since harmonic drive with high gear ratio is used in each joint to transmit the torque from the motor to the link, the independent joint controller can reject the effect of the nonlinear disturbance by utilizing cascaded control system. Thus the proposed independent joint controller consists of a inner velocity loop which is cascaded with a outer position loop. For the inner loop a PI controller is used while for the outer loop P controller is employed. The controller gains are tuned based on the step response for each joint motor. In order to imitate the human walking, the three dimensional trajectories of the feet and the torso are developed. For constructing the trajectories, first, motion constraints during walking are derived based on the analysis of human walking pattern. Then cubic spline interpolation is used to find the smooth trajectories for the feet and the torso in both single and double-support phase. The trajectories generated by walking pattern generator can be redesigned easily by changing the walking parameters. The closed-loop solution of inverse kinematics is developed to convert the desired trajectories from the operational space to the joint space. The closed-loop solution of the inverse kinematics is superior with respect to the iterative solution due to the less computation time. In addition, a kinematic simulation is developed to illustrate the robot configuration before implementation. For implementation, a C++ program is developed to generate the reference joint angle trajectories. This program convert these trajectories to digital number and put them in Position-Time (PT) table. These data are sent to each joint controller under CAN message format. In this manner, the controller actuate the motor by generating proper voltage to synchronize the motion of the robot joints. In order to realize the biped walking in the sense of static stability, the robot's center of gravity should be located on the above of the support foot area. Therefore, many experiments have been done to find the optimal values of walking parameters. Finally, stable walking realized for the biped robot with speed up to 0.076 km/h.