Myoelectric signals from several muscles of the lower limb were studied under various speed and stride length conditions. The main purpose was to determine invariant and variant features among these myoelectric patterns. A pattern recognition algorithm was used to analyze these activity patterns. Within-condition analysis revealed some common features among the EMG patterns. This suggests that the nervous system does not have to generate all the muscle activity patterns, only the common features that can, in appropriate combination, produce the necessary activity patterns. From the across condition analysis, the following rules emerged. First, both phasic component and magnitude (d.c. level) of the muscle activity patterns have to be modulated to meet the demands imposed by the various conditions. Second, the variability in the proximal muscle activity patterns across conditions are higher than the distal muscle activity patterns. Within each group, the extensor muscles and double-jointed muscles show greater variability than the flexor muscles and single-jointed muscles. And finally, the changes in the average value (d.c. level) of the muscle activity patterns across conditions are not uniform but show muscle and task specificity. For example, within the speed condition, the increase in d.c. level of the extensors with speed of locomotion show a proximal to distal trend. Based on these results, a conceptual model for the human locomotor control process is proposed.