Memories depend on persistent changes in transmission at synapses in the brain. The idea seems easy to accept now but Lashley, for instance, once wrote ‘Among the many unsubstantiated beliefs concerning the physiology of the learning process, none is more widely prevalent than the doctrine that the passage of the nerve impulse through the synapse somehow reduces synaptic resistance and leads to the fixation of a new habit’. In the two papers revisited here, Tim Bliss, Terje Lomo and Tony Gardner-Medwin (Bliss & Lomo, 1973; Bliss & Gardner-Medwin, 1973) substantiated one of those beliefs by demonstrating persistent, activity-dependent changes at hippocampal synapses. Several ideas on memory formation were current when the work was initiated. Ramon y Cajal had proposed that neuronal growth was the substrate for memory. Lashley suggested that memories were not stored locally but distributed engrammatically throughout the cortex. Molecular studies had begun with the demonstration of changes in mRNA of Deiters nucleus cells after learning (Hyden & Egyhazi, 1962). The idea that firing could reverberate in closed chains of neurones, originally envisaged by Lorente de No, was being explored in the cortex by Burns (1957). Hebb was also attracted to the idea that reverberating activity might preserve new memories until growth could occur. He had postulated (1949) that transmission in recurrent circuits might be persistently modified by coincident activity at single connections. However, while persistent changes in efficacy had been seen at single invertebrate synapses (Kandel & Tauc, 1965), long-term changes had not been seen in mammals before 1973. The work of Bliss, Lomo and Gardner-Medwin focused on the hippocampus, an area associated with memory formation by work on the patient HM (Scoville & Milner, 1957). They examined transmission at perforant path synapses made with hippocampal dentate granule cells. One study was done on anaesthetized rabbits and the other on animals chronically implanted with recording electrodes. Both papers relied on interpretation of hippocampal field potentials developed by Per Andersen, the PhD supervisor of Terje Lomo who also welcomed Tim Bliss to Oslo as a postdoc. Records of extracellular responses to perforant path stimuli let them study the population EPSP, an index of synaptic transmission, and the population spike which reflected synaptically induced firing. Both the population EPSP and the population spike were increased, for hours in anaesthetized animals and for days in implanted rabbits, after high frequency stimuli. Thus two phenomena were born: long-term potentiation (LTP) of synaptic transmission and of the coupling between EPSP and spike initiation. Reading the papers again reminds us that the first long-term changes were highly variable. The population spike was sometimes potentiated with no change in the population EPSP and vice versa. In the implanted animals, a potentiation lasting more than 1 h was seen in only 26% of trials. Also the discussions were remarkably prescient, identifying a number of potential mechanisms. We examine in the rest of this perspective, how the unravelling of these mechanisms has provided a cell biological explanation of one type of synaptic plasticity and hugely contributed to our understanding of neuronal physiology. In 1973 neither the receptors nor the neurotransmitter at hippocampal excitatory synapses were identified. Even without this information, Bliss and Lomo proposed several mechanisms for an increase in synaptic efficacy: (i) an increased number of terminals releasing transmitter, (ii) an increased amount of released transmitter, (iii) a reduced resistance of spine necks, or (iv) an increase in the sensitivity of the postsynaptic junctional membrane. They provided a clear prospectus for future work. The phenomenon of LTP was confirmed in the then novel slice preparation (Schwartzkroin & Wester, 1975) and shown to depend on the activation of synaptic receptors (Dunwiddie et al. 1978). The emergence of an excitatory amino-acid pharmacology in the early 80 s (Davies et al. 1981) permitted identification of NMDA receptors as a key element in LTP induction (Collingridge et al. 1983). The pivotal role of postsynaptic Ca2+ entry was demonstrated by Lynch et al. (1983). An increase in glutamate binding after LTP suggested that the Ca2+ elevation might lead to synaptic potentiation by increasing the density of postsynaptic receptors rather than their sensitivity (Lynch & Baudry, 1984). While this model resembles some current views of LTP expression, it took some years for some sort of consensus to emerge. Patch-clamp techniques (Edwards et al. 1989) provided another tool to discriminate between pre- and postsynaptic sites for LTP expression. A quantal analysis of variations in synaptic responses should have resolved the question. Instead it increased the confusion. Different studies demonstrated an increase in postsynaptic responses (Foster & McNaughton, 1991) in transmitter liberation (Malinow & Tsien, 1990), or both (Kullmann & Nicoll, 1992). The contradiction was resolved by suggesting that LTP uncovers silent synapses, recruiting newly functional AMPA receptors to sites previously expressing only NMDA receptors (Kullmann, 1994). The mechanism is postsynaptic but paradoxically the number of effective release sites is increased (Malinow & Malenka, 2002). To understand LTP, we must now understand receptor trafficking. Yet this model of LTP expression does not fit all synapses. LTP at mossy fibre connections with CA3 pyramidal cells works differently (Nicoll & Malenka, 1995). Even at the Schaffer collateral synapse with CA1 pyramidal cells there is evidence for a presynaptic component to long-term changes (Emptage et al. 2003). Furthermore, most studies on LTP are quite short-lasting – compared with the duration of an animal, or human, memory. Although uncovering silent synapses may explain the early phase of LTP, later components probably involve additional mechanisms including protein synthesis (Duffy et al. 1981) and physical changes in synaptic structure (Toni et al. 1999). The second phenomenon discovered by Bliss, Lomo and Gardner-Medwin was that the ability of EPSPs to discharge a cell could be persistently enhanced. Their key evidence was that population spike amplitudes could increase even when population EPSPs were unchanged by tetanic stimuli. They concluded ‘potentiation of the spike parameters could not be explained wholly in terms of potentiation of the EPSP’. Later work (Andersen et al. 1980) confirmed changes in the relationship between EPSP and population spike amplitudes pointing to a distinct cellular rather than synaptic plasticity. Thus was launched, with longer latency, a second wave of studies on persistent changes in cellular excitability. Bliss and Lomo noted that an increased population spike amplitude might imply either more cells firing or an increased synchrony of their discharge. They proposed two mechanisms: a reduction in the efficacy of synaptic inhibition or an increase in the excitability of the postsynaptic cell. The hypothesis of a reduced efficacy in inhibitory synaptic circuits is supported by occlusion experiments showing that blocking GABA receptors suppressed EPSP-spike potentiation (Abraham et al. 1987). More recent work has identified a pathway by which Ca2+ entry activates the Ca2+-dependent phosphatase 2B, depresses GABAergic signalling and so facilitates EPSP-spike coupling (Lu et al. 2000). EPSP-spike potentiation might also result from an increase in cellular excitability, best examined by testing action potential generation by a single cell. Single cell studies, which began as late as 1990 (Chavez-Noriega et al. 1990), suggest that cellular excitability as well as synaptic efficacy can be persistently altered (Daoudal et al. 2002). The diversity and versatility of voltage gated channels expressed by neurones is far richer than imagined in 1973. Patterned synaptic stimulation has been shown to induce maintained changes in several cellular currents (Turrigiano et al. 1994; Aizenman & Linden, 2000; Frick et al. 2004). Reciprocally, persistent changes in cellular excitability can alter synaptic efficacy homeostatically so as to stabilize firing (Burrone et al. 2002). So while Bliss, Lomo and Gardner-Medwin showed how the brain might change, now we search for ways to ensure that it remains the same. After all this work, have we made any progress with the other unsubstantiated belief that troubled Lashley? Can persistent changes in synaptic efficacy or cellular excitability fix a new habit? Changes in synaptic function in reflex circuits underlying behaviours such as olfactory learning and acquisition of fear responses seem to be correlated with learning an olfactory task (Roman et al. 1993) or fear conditioning (Rogan et al. 1997). Furthermore interference with molecules thought to participate in long-term plasticity block some forms of learning (Tsien et al. 1996; Mansuy et al. 1998). However, the most difficult problem remains to be solved. The trouble is coding. How do persistent changes in efficacy at single synapses or in the excitability of single neurones contribute to memorizing a fact or remembering a place (Lever et al. 2002)? Are memories not linked to some kind of ensemble activities, and how does recall work (Gardner-Medwin, 1976)? These issues must still be addressed. Even so we have come a long way. The 1973 papers from Bliss, Lomo and Gardner-Medwin have been hugely stimulating for a generation of neurobiologists. They tell their side of the story at –http://www.ergito.com/main.jsp?bcs=EXP.13.8– we should be grateful to them.