Activity exerts powerful, selective effects on the organisation, stability and strength of synaptic connections in all parts of the nervous system, including neuromuscular junctions (Lohof et al. 1996;Sanes & Lichtman, 1999). For example, neonatal rat muscle fibres are innervated at single motor endplates by motor nerve terminals supplied by several motoneurones (polyneuronal innervation; ð_junctions), and a similar pattern is re-established in adult muscle after nerve injury and regeneration (Brown et al. 1976;Betz et al. 1979;Ribchester, 1988). With time, synaptic boutons and axonal inputs are progressively eliminated, eventually leaving most endplates innervated by only one motor axon (Gan & Lichtman, 1998). Progressive and disproportionate weakening of synaptic transmission, induced by differences in pre-and postsynaptic activity, has been presumed to underlie this competitive process (Ribchester & Taxt, 1983;Balice-Gordon & Lichtman, 1994;Coleman et al. 1997). However, activity may not be sufficient to induce elimination from all ð_junctions. Previous studies of paralysed neonatal or reinnervated adult muscle suggest that significant polyneuronal innervation persists once activity resumes (Hoffman, 1953;Brown et al. 1982;Barry & Ribchester, 1995). The properties that allow convergent synapses of different size and efficacy to persist at a motor endplate are unknown. Equally, it is unclear whether apparently stable, convergent synaptic inputs in reinnervated muscles may in