Electrical stimulation offers the possibility of restoring motor function of paralyzed limbs after spinal-cord injury or stroke, but few data are available to compare possible sites of stimulation, such as muscle, nerve, spinal roots, or spinal cord. The aim of this study was to establish some characteristics of stimulation at these sites in the anesthetized and midcollicular decerebrate cat. The hind limb was constrained to move in the sagittal plane against a spring load. Ventral-root stimulation only produced movements down and back; the direction moved systematically backward the more caudal the stimulated roots. In contrast, dorsal-root stimulation only produced movements up and forward. Thus, neither method alone could produce the full range of normal movements. Muscle, nerve, and intraspinal stimulation within the intermediate regions of the gray matter generated discrete, selective movements in a wide range of directions. Muscle stimulation required an order of magnitude more current. Single microwire electrodes located in the spinal gray matter could activate a synergistic group of muscles, and generally had graded recruitment curves, but the direction of movement occasionally changed abruptly as stimulus strength increased. Nerve stimulation produced the largest movements against the spring load (>80% of the passive range of motion) and was the most reproducible from animal to animal. However, recruitment curves with nerve stimulation were quite steep, so fine control of movement might be difficult. The muscle, nerve, and spinal cord all seem to be feasible sites to restore motor function. The pros and cons from this study may be helpful in deciding the best site for a particular application, but further tests are needed in the chronically transected spinal cord to assess the applicability of these results to human patients.