It is here proposed that the evanescent network derived from malleable or supple synapses is the substrate for long-lasting memory. The subjective sense of memory permanence is not derived, as suggested by Bain and others, from the stabilization of synaptic structure which gives rise to consolidated distributed networks. This generally held wisdom that synapses are activated and ultimately stabilized to reflect the long-lasting substrate of memory is reinforced by increased interest in the importance of sparse coding in memory consolidation. One line of evidence for sparse coding derives from studies on the lateral nucleus neurons of the amygdala (for review, see Josselyn, 2010). These findings lead to the conclusion that a small number of neurons are both necessary and sufficient to retrieve the fear engram. Recently, it has been shown that sparse coding in the dentate gyrus of the hippocampus is sufficient for retrieving the fear conditioning engram (Liu et al., 2012). One implication of these findings is that memory is stored in selected cells and that this restricted storage is more or less permanent, as the authors note: “… Defined cell populations can form the cellular basis for fear memory engrams (Liu et al., 2012, p. 89).” But the problem with this model is that because new learning is incorporated into existing networks, stabilization would work against this integration. For this reason and because of obligatory processes of metabolic protein turnover and ongoing synaptic malleability, the “supple synapse” model is proposed (Routtenberg, A., Rekart, 2005; Routtenberg, 2008a,b). Specifically, long-lasting memory is derived from the instructive, use-dependent sampling of neural networks selected from a very large universe of networks that are evanescent because they are linked together by supple synapses. Importantly, if such suppleness did not exist, the stabilization of synapses would then prevent the physiological malleability of brain circuitry that is essential both for proper integrated information storage and for flexible information retrieval. A corollary of this proposal is that it suggests an alternative view of consolidation: the same agent which is disruptive immediately after learning is no longer effective later because the network, over time, becomes widely distributed and evanescent. Thus, time-dependence is replaced with space-dependence. © 2013 Wiley Periodicals, Inc.