Mechanisms of long-term potentiation (LTP) maintenance are discussed in the light of the phenomenon of silent synapses. Evidence that LTP is associated with the insertion of new AMPA receptors (AMPARs) in postsynaptically silent (deaf) synapses expressing only NMDA receptors (NMDARs) before LTP induction has led to the assumption that the debate on pre- versus postsynaptic locus of LTP expression has been resolved in favour of the latter. However, recent data indicate that these synapses are mainly presynaptically silent (mute or whispering), because the probability of glutamate release (P(r)) or glutamate concentration in the cleft is too low to activate AMPARs. In this case LTP could be explained by an increase in P(r) or enhanced glutamate concentration to activate low affinity AMPARs. Optical methods to probe calcium transients in dendritic spines have revealed an increase in P(r) during LTP with concomitant postsynaptic modifications. A hypothesis is considered that accounts for the differences in both the initial failure rates between AMPAR- and NMDAR-mediated responses, and the LTP-associated decrease in failures of AMPAR-mediated responses. According to this hypothesis, glutamate release is potentiated by the strong postsynaptic depolarization used to identify NMDAR-mediated responses. We suggest that the expression of LTP may depend on coordinated pre- and postsynaptic modifications whose relative contributions vary according to the initial state of the synapse, the experimental protocol and time after induction.