The neuromuscular junction undergoes a loss of synaptic connections during early development. This loss converts the innervation of each muscle fiber from polyneuronal to single. During this change the number of motor neurons remains constant but the number of muscle fibers innervated by each motor neuron is reduced. Evidence indicates that a local competition among the inputs on each muscle fiber determines which inputs are eliminated. The role of synapse elimination in the development of neuromuscular circuits, other than ensuring a single innervation of each fiber, is unclear. Most evidence suggests that the elimination plays little or no role in correcting for errant connections. Rather, it seems that connections are initially highly specific, in terms of both which motor neurons connect to which muscles and which neurons connect to which particular fibers within these muscles. A number of attempts have been made to determine the importance of neuromuscular activity during early development for this rearrangement of synaptic connections. Experiments reducing neuromuscular activity by muscle tenotomy, deafferentation and spinal cord section, block of nerve impulse conduction with tetrodotoxin, and the use of postsynaptic and presynaptic blocking agents have all shown that normal activity is required for normal synapse elimination. Most experiments in which complete muscle paralysis has been achieved show that activity may be essential for the occurrence of synapse elimination. Furthermore, experiments in which neuromuscular activity has been augmented by external stimulation show that synapse elimination is accelerated. A plausible hypothesis to explain the activity dependence of neuromuscular synapse elimination is that a neuromuscular trophic agent is produced by the muscle fibers and that this production is controlled by muscle-fiber activity. The terminals on each fiber compete for the substance produced by that fiber. Inactive fibers produce large quantities of this substance; on the other hand, muscle activity suppresses the level of synthesis of this agent to the point where only a single synaptic terminal can be maintained. Inactive muscle fibers would be expected to be able to maintain more nerve terminals. The attractiveness of this scheme is that it provides a simple feedback mechanism to ensure that each fiber retains a single effective input.