This article reports two series of investigations into the formation of synapses after injury in the adult rat. A comparison is made between synapse formation in a peripheral nervous site - the superior cervical sympathetic ganglion after destroying almost all the afferent fibres by cutting the preganglionic chain, and a central nervous site - the septal nuclei, after selectively destroying afferent fibres from the hippocampus running in the fimbria. The method consists of a quantitative electron microscopic assessment of numbers of synapses in the neuropil of the deafferented regions at progressively increasing survival times after the injury. The superior cervical ganglion has about 8.8 x 10<latex>$^6$</latex> synapses and about 8000 preganglionic axons. After a freeze lesion of the preganglionic chain over 90% of the synapses disappear, indicating that the preganglionic axons each form an average of about 1100 synaptic contacts. A proportion of the deafferented postsynaptic sites are recognizable by the presence of the vacated postsynaptic thickenings. As the preganglionic axons regenerate back across the lesion, the synapses gradually reappear, and the vacated thickenings disappear. If an insufficient number of axons regenerates, less than the normal complement of synapses is formed; the number of synapses bears a constant relation to the number of axons, and under these circumstances each axon forms about the same number of synapses as in the normal ganglion. This suggests that the number of synapses which an axon can form is limited, and that the preganglionic axons may be close to this limit in the normal ganglion. The ganglion can also be reinnervated by foreign nerves such as the vagus and hypoglossal. These nerves form less than normal numbers of synapses. In all cases of less than complete reinnervation, the numbers of vacated thickenings remaining are inversely proportional to the numbers of synapses which appear; this indicates that the synapses are formed at the originally denervated postsynaptic sites. The septum receives inputs of hippocampal origin both through the ipsilateral and the contralateral fimbria. After section of one fimbria, the terminals of the cut axons are recognizable by the reaction of electron dense degeneration. About half the synapses on dendritic spines belong to the ipsilateral fimbria and about half this number to the contralateral fimbria. The degenerating terminals are progressively removed from the postsynaptic sites by the phagocytic action of astrocytes. However, unlike the ganglion, the sites only rarely appear as vacated thickenings. Instead, there is a progressive increase in the number of non-degenerating terminals as the degenerating terminals are removed. This spontaneous reinnervation process results in restoration of the original complement of synapses. However, the originally cut fimbrial axons do not regenerate: this process of synapse formation is due to formation of additional contacts by the other axons remaining in the deafferented region. If the second fimbria is now cut, it results in an amount of degeneration equal to the sum of that found after cutting both the ipsilateral and the contralateral fimbria. This indicates that the sites left vacant by section of one fimbria are reinnervated exclusively by the other fimbria. However, this preference is not absolute. After section of the second fimbria, the degeneration is still removed, and a normal complement of synapses once again restored. Thus the septal neuropil has great powers of re-establishing normal synapse numbers even after major deafferentation. The ganglion and the septum are similar to the extent that denervated postsynaptic sites are capable of being reinnervated - spontaneously in the case of the central nervous site. The major difference is that in the peripheral nervous site the originally cut axons can regenerate back to their former targets. It is not clear why the cut fimbria - behaving in a manner typical for central nervous tracts - does not regenerate. The proximal parts of the cut fimbrial axons and their cells of origin survive the injury. The denervated sites in the septum also survive, and are capable of being reinnervated. Further, this reinnervation process is selective for fimbrial axons from the opposite hippocampus. This also shows that fimbrial fibres are capable of forming new synapses. Thus the defect which prevents true regeneration of the originally cut axons back to their former targets does not appear to be due to failure in the mechanism of synapse formation or matching. This suggests that the defect may be an inability of the cut axons to regenerate across the site of the lesion.