The multiterminal slow and fast neuromuscular junctions of the moth Manduca sexta were compared using scanning, thin-section, and freeze- fracture techniques to see what structural features might underlie their functional differences. Slow neuromuscular junctions, here formed on tonic muscle fibers, produce a facilitating e.j.p. the amplitude of which is 1/5 to 1/3 the size of a fast excitatory junction potential (EJP) and the duration of which is nearly four times longer. A slow junction consists of a single terminal branch that is shorter in length than either of the pair of branches that a fast junction forms close together on the muscle fiber. Within the junction, slow nerve terminals exhibit longer, more frequent constrictions and are very varicose compared with fast. Since fast larval junctions on tonic muscle fibers are also varicose (Schaner and Rheuben, 1985), this is unlikely to represent an intrinsic property of the nerve. However, calculations of the length constants of the varicose versus nonvaricose shapes indicate that the effect of passive cable properties on normal functioning may act to limit the length of the slow terminals more than that of fast. Even though the varicose shape can be predicted to prolong the time course of the EJP, calculations show that, at the measured length, this would not explain the very long EJP that is observed. Within the neuromuscular junctions, the synapses are characterized on the muscle membrane by a patch of densely packed particles on the external leaflet and on the nerve membrane by a single linear active zone. The total number of synapses per slow junction is about 1/3 that of fast junctions. There is a weak correlation between average area of the individual postsynaptic particle patches and cross-sectional area of the muscle fibers that transcends nerve and muscle fiber types. The average lengths of active zones from the two types do not differ significantly. However, the number of particles per active zone in slow junctions is about 55% of the number in fast active zones. Chemically fixed slow nerve terminals have a greater density of synaptic vesicles remaining than do fast. If a proportion of the active zone particles represent structures directly involved in the probability of transmitter release, such as Ca++ channels, then the latter two characteristics may jointly reflect differences in capability to release and mobilize transmitter that would partly explain the different EJP amplitude and facilitation properties.