LTD of synaptic transmission is a form of long-term synaptic plasticity with the potential to be as significant as LTP to both the activity-dependent development of neural circuitry and adult memory storage. In addition, interactions between LTP and LTD and the dynamic regulation of the gain of synaptic plasticity mechanisms are also very important. In particular, the computational ability of LTD to properly counterbalance LTP may be essential to maintaining synaptic strengths in the linear range, and to maximally sharpen the ability of synapses to compute and store frequency-based information about the phase relation between synapses. Experimental data confirm the presence of an activity-dependent "sliding threshold" with the expected properties. That is, when levels of neuronal activity are high, indicating circumstances increasing the likelihood of inducing LTP, compensatory changes cause the suppression of LTP and an enhanced likelihood of LTD. Conversely, we would predict that low levels of synaptic activity would shift the threshold in favor of greater LTP and less LTD, a hypothesis which has yet to be tested. The sliding threshold for LTP and LTD also has implications for underlying cellular mechanisms of both forms of long-term synaptic plasticity. If the thresholds for LTP and LTD are tightly and reciprocally co-regulated, that could imply that at least one component of LTD is a true depotentiation caused by reversal of a change mediating LTP. If so, the intuitively simplest hypothesis is that phosphorylation of AMPA glutamate receptors causes LTP of synaptic e.p.s.p.s, while dephosphorylation of the same site or sites causes depotentiation LTD. Of course, this hypothesis would refer only to a postsynaptic component of both LTP and LTD. There has been a recent report that, in neonatal rat hippocampus, a form of LTD that is expressed developmentally earlier than LTP appears to have a postsynaptic induction site, but is expressed as decreased presynaptic transmitter release (Bolshakov and Siegelbaum, 1994). Whether these properties will be retained as LTD matures is unknown, as is the likelihood that, if a component of LTP is expressed presynaptically, depotentiation of that presynaptic component can also occur. Equally unclear is the persistence of LTD relative to LTP. The few rigorous long-term anatomical studies available suggest that the latest phases of LTP may be expressed as changes in dendritic spine shapes and/or synaptic morphology. While heterosynaptic LTD has been reported to have a duration of weeks in vivo (Abraham et al., 1994), we do not know whether LTP-induced morphological changes that take many days to appear can be reversed in an activity-dependent manner. An important feature of the consolidation of memories may turn out to be the slow development of LTP that is resistant to reversal by LTD. While we still at an earlier stage in our understanding of the mechanisms underlying LTD compared to LTP, some things are becoming clear. LTD is induced by afferent neuronal activity that is relatively ineffective in exciting the postsynaptic cell--an "anti-hebbian" condition. This property, coupled with the hebbian properties of LTP and the dynamic nature of membrane conductances, necessarily confers upon synapses the ability to compute and store the results of a covariance function. However, the role of such a computation in processing and/or memory is unclear. In addition, LTD appears to require the activation of NMDA and metabotropic subtypes of glutamate receptors, release of Ca2+ from intracellular stores, and an increase in intracellular [Ca2+] that is lower than that necessary to induce LTP. The early evidence is consistent with some overlap of targets for modification by LTP and LTD, with some forms of LTD likely to be a reversal, or "depotentiation," of previous LTP, perhaps through dephosphorylation of AMPA receptors.