Long-term potentiation (LTP), a model of activity-dependent synaptic plasticity and of certain forms of memory, comprises the persistent enhancement of excitatory neurotransmission that results from high-frequency activation. A presynaptic component of LTP is thought to be modulated by a retrograde messenger generated by the postsynaptic neuron. Arachidonic acid, nitric oxide, carbon monoxide and PAF have each been proposed as retrograde messengers in LTP, but arachidonic acid, unlike PAF, requires NMDA receptor activation. A PAF antagonist (BN 52021) that provides neuroprotection in ischemia-reperfusion displaces [3H] PAF bound to presynaptic membranes, blocks PAF-induced glutamate exocytosis and inhibits LTP. An antagonist selective for the intracellular PAF binding site (BN 50730) did not affect LTP, nor did BN 52021 modify NMDA currents. LTP was induced with weak synaptic stimulation coupled with postsynaptically administered enzyme resistant mcPAF. Theta-burst stimulation (10 min) after bath applications of mcPAF (1 microM) induced APV-independent LTP that was blocked by 5 microM BN 52021. When this antagonist was infused into the hippocampus before or immediately after training, it impaired memory of inhibitory avoidance training in the rat. Memory was not altered if the antagonist is infused 30 or 60 min after training. Moreover, mcPAF enhances memory on retention test performance of step-down inhibitory avoidance habituation and learning in rats. Also, memory was studied using a caudate nucleus-dependent cued water maze task. Rats received an 8 trial (30 s intertrial interval) training session in which a visible cued escape platform was located in a different quadrant of the maze of each trial. Following trial 8, the rats received a unilateral post-training intra-caudate injection of mcPAF (1 microgram/0.5 microliter), BN 52021 (0.5 microgram/0.5 microliter) or vehicle. On a retention test session 24 h later, latency to mount the escape platform was used as a measure of memory. The retention test escape latencies of rats given mcPAF were significantly lower than those of the vehicle-injected controls, indicating a memory enhancing effect of mcPAF. Injection of mcPAF did not affect retention when administered 2 h post-training, indicating a time-dependent effect of mcPAF on memory. The latencies for animals injected with BN 52021 were significantly higher than those of the controls, indicating that antagonism of endogenous PAF impairs memory. The findings show that PAF plays a role in memory formation in a caudate-mediated cued discrimination task. Administration of BN 52021 2 h post-training had no affect on retention, indicating a time-dependent effect of endogenous PAF on memory formation. PAF, the most potent bioactive lipid known, modulates excitatory synaptic transmission, neuronal plasticity and memory. When PAF production is overstimulated as in seizures or ischemia, it becomes neurotoxic.