The prevailing textbook view that de novo protein synthesis is required for memory (e.g., [Bear, M. F., Connors, B., & Paradiso, M. 2006. Neuroscience. Lippincott, New York]) is seriously flawed and an alternative hypothesis has been proposed in which post-translational modification (PTM) of proteins already synthesized and already present within the synapse is 'the' substrate for long-lasting memory. Protein synthesis serves a replenishment role. The first part of this review discusses how long-lasting memory can be achieved with 'only' PTM of existing synaptic proteins. The second part critically reviews a recent report published in Neuron 2007 that exemplifies the current view of protein synthesis and memory while also illustrating how these results can be understood within this new PTM framework. A necessary yet unexpected conclusion to emerge from consideration of the consequences of a PTM mechanism as the necessary, sufficient and exclusive substrate for long-lasting memory, is that the central Hebbian dogma that cells that 'fire together, wire together' is an unlikely mechanism for long-lasting memory. Thus, a unique feature of the PTM model is that longevity of information storage is achieved not by stability of the synaptic mechanism, but by impermanent pseudoredundant circuits. This is so because PTM is a reversible process and thus any permanent connection, any 'lasting effect' cannot be in the form of stable synapse formation. We have therefore proposed a solution in which network level processes regulate cellular mechanisms, even as such mechanisms regulate the network. Thus, synapses are 'meta-stabilized' by regulated feedback mediated by the circuit in which the synapse is embedded. For example, spontaneous activity is proposed to be a substrate feedback mechanism we term 'cryptic rehearsal' to sustain for some period of time after learning an approximation to the state initially created by input. Additionally, because the duplication of these traces is ongoing, this provides a degenerate code for the engram. Stability is thus achieved, not by stabilizing the synapse, but by implementing a pseudo-redundant yet malleable circuitry so that memory can be protected in the face of small catastrophes in network representation.