Learning and memory formation are apparently based on cascades of mole
cular and cellular processes with increasing time constants (ms to day
s and weeks), but even the most long-lasting effects are transient. Me
mory traces may permanently modify the behavior (activity patterns, ge
ne expression) of neurons and neuronal networks. Therefore the questio
n is raised whether our current view on the stability of synapses unde
r normal conditions is tenable. Evidence is reviewed suggesting that a
s direct or indirect effects of modifications in bioelectrical activit
y and chemical trophicity, synapses may be remodeled and removed withi
n days and weeks, and possibly within hours. Accordingly, species-spec
ific connectivity patterns are not restricted to the standard architec
ture of the CNS, but (morpho-)genetics allow for a considerable number
of alternative wiring patterns, which appear under unusual conditions
during ontogenesis and in adulthood. Our present knowledge suggests t
hat, rather than the formation of synapses, they are a selective proce
ss. Until now there is no direct method of measuring either synaptic r
eorganization or the average life span of synapses. Specific cases, ho
wever, allow to estimate synapse turnover during ontogenesis, at its l
owest possible level. Such data suggest that each synapse is on averag
e remodeled or replaced several to many times during normal developmen
tal, e.g. in the cerebral cortex of Marmoset monkeys at the very least
5 to 10 times (corresponding to 250 million synapses eliminated per h
our in area 17!). It is discussed how the consequences of synapse turn
over could be utilized by learning processes. Conclusions are followed
by an outlook.