The view on the neural control of human locomotion has undergone changes in the past decades. In spite of the encephalization and the erect, bipedal mode of walking characteristic for human beings, independent observations imply that unperturbed locomotor patterns can be generated by similar spinal neural circuits as in other vertebrates. However, little is known about the organization of these rhythm and pattern generating networks in humans. It has been shown that the human lumbar spinal cord isolated from supraspinal control due to traumatic spinal cord injury (SCI) can generate rhythmic, locomotor-like activity in response to sustained epidural spinal cord stimulation (SCS) of certain frequencies. The rhythmic activities consist of a series of stimulus time-related and rhythmically modulated posterior-root muscle (PRM) reflexes, each initiated in posterior root afferents and electromyographically recorded as compound muscle action potentials (CMAPs). The relation between individual stimuli and responses, as well as their electromyographic (EMG) characteristics, allow for the identification of mechanisms in addition to the information gained from the overall EMG patterns. This thesis aims at uncovering the locomotor capabilities and their underlying mechanisms intrinsic to the human lumbar spinal cord. In the first part, rhythmic EMG data in response to SCS were analyzed, both, regarding overall patterns and the constituent units. Based on the information gained, in the second part, computer models were formulated to test hypotheses, to learn about their implications to primary and secondary phenomena and to generate research questions. EMG activities of quadriceps, hamstrings, tibialis anterior and triceps surae, bilaterally in response to epidural stimulation at - 42 Hz were analyzed in 10 individuals with motor complete posttraumatic SCI. Thirty-nine segments (duration: 10 s) of rhythmic activities found in all four-muscle groups of one lower limb were identified in 7 subjects. Phases of bursting and suppressed activities were recognized. Latencies of PRM reflexes were calculated. A computational network model of neurons with Hodgkin-Huxleylike membrane dynamics was developed to test whether hypothesized rhythm and pattern generating networks would reproduce the recorded data. A core rhythm-generating network model was extended by adding conduction delays, presynaptic inhibition and disinhibition of parallel central pathways. Within a given 10-s segment, rhythmic activities of all muscle groups had a constant phase relation. Dimensionality reduction by non-negative matrix factorization revealed that all expressed activity patterns of individual muscle groups can be best reproduced by a linear combination of 3 to 4 basic patterns, while two basic patterns that are similar to those seen in fictive locomotion (co- and reciprocal activity) already explain 83.2% of the variance. PRM reflexes constituting bursts during the extension phases had predominantly monosynaptic latencies. During flexion phase, a suppression of these responses was often observed. In such cases these monosynaptic reflexes were replaced by delayed, oligosynaptic PRM reflexes. Computer simulation showed that the activation of the rhythm and pattern generating circuits with persistent sodium channels as the source for rhythm generation is frequency dependent and this frequency dependence matches the electrophysiological data beyond the hypotheses. Stimulus time-locked motoneuron firing (resulting in the PRM reflexes) was explained by the interplay of relatively strong and highly synchronized afferent input and the relatively diffuse and weaker influence of the interneurons. This afferent influence of the motoneurons was presynaptically, rhythmically gated and postsynaptically modulated by excitatory input from the pattern formation and inhibitory input from last-order interneurons. Together with the selection of alternative interneuronal pathways (within the flexor side) the model reliably reproduced electrophysiological findings. The electrophysiological data, as well as the computer simulations, give insight into the organization of the human spinal rhythm and pattern generating networks and reveal common control characteristics with the central pattern generators for locomotion described in animal experimental work. The constant phase relation of rhythmic outputs to one lower limb suggests a common, plurisegmental rhythm generator, and the various EMG patterns indicate separate stereotypic pattern formation modules. These neural circuitries possess many of the necessary components to generate functional locomotion. Yet, there is a lack of coordination in and between the muscles. Such coordination may require inputs from supraspinal centers, as well as feedback from the periphery.