Glutamate is the most widespread excitatory transmitter in the CNS and is probably involved in LTP, a neural phenomenon which may be associated with learning and memory formation. Intracerebral injection of large amounts of glutamate between 5 min and 2.5 min after passive avoidance learning in young chicks inhibits short-term memory, which occurs between 0 and 10 min post-learning in a three-stage model of memory formation first established by Gibbs and Ng(25) [Physiol. Behav. 23:369-375; 1979]. This effect may be attributed to non-specific excitation. Blockade of glutamate uptake by L-aspartic and beta-hydroxamate also abolishes this stage of memory, provided the drug is administered within 2.5 min of learning. Interference with either production of percursors for transmitter glutamate in astrocytes or with glutamate receptors is also detrimental to memory formation, but the effects appear much later. After its release from glutamatergic neurons, glutamate is, to a large extent, accumulated into astrocytes where it is converted to glutamine, which can be returned to glutamatergic neurons and reutilized for synthesis of transmitter glutamate, and partly oxidized as a metabolic substrate. The latter process leads to a net loss of transmitter glutamate which can be compensated for by de novo synthesis of a glutamate precursor alpha-ketoglutarate (alpha KG) in astrocytes, a process which is inhibited by the astrocyte-specific toxin fluoroacetate (R. A. Swanson, personal communication). Intracerebral injection of this toxin abolishes memory during an intermediate stage of memory processing occurring between 20 and 30 min post-training (50) [Cog. Brain Res, 2:93-102; 1994]. Injection of methionine sulfoximine (MSO), a specific inhibitor of glutamine synthetase, which interferes with the re-supply of transmitter glutamate to neurons by inhibition of glutamine synthesis in astrocytes, has a similar effect. This effect of MSO is prevented by intracerebral injection of glutamate, glutamine, or a combination and alpha KG and alanine. MSO must be administered before learning, but does not interfere with acquisition since short-term memory remains intact. Administration of either the NMDA antagonist AP5, the AMPA antagonist DNQX, or the metabotropic receptor antagonist MCPF, also induces amnesia. Memory loss in each case does not occur until after 70 min post-training, during a protein synthesis-dependent long-term memory stage which begins at 60 min following learning. However, to be effective, AP5 must be administered within 60 s following learning, MCPG before 15 min post-learning, and DNQX between 15 and 25 min after learning. Together, these findings suggest that learning results in an immediate release of glutamate, followed by a secondary release of this transmitter at later stages of processing of the memory trace, and that one or both of these increases in extracellular glutamate concentration are essential for the consolidation of long-term memory. Since both fluoroacetate and MSO act exclusively on glial cells, the findings also show that neuronal-glial interactions are necessary during the establishment of memory.