Spatial pretraining can enable spatial learning in another environment that ordinarily requires hippocampal N-methyl-d-aspartate (NMDA) receptor activity to become independent of that activity. This study explored further the circumstances in which this training-induced ‘rescue’ of later learning in the presence of the NMDA receptor antagonist 2-amino-5-phosphonovaleric acid (D-AP5) can occur. D-AP5 (0, 10, 20 and 30 mm in artificial cerebrospinal fluid) was infused continuously (0.5 μL/h, from a minipump) and bilaterally into the dorsal hippocampus during spatial-reference-memory training in a watermaze (4 trials/day, 8 days). This was preceded either by handling only or by identical spatial training in another watermaze in a separate laboratory with different extramaze cues. In naïve rats, D-AP5 caused a dose-related impairment in spatial reference memory acquisition that was significant at the lowest 5 nm/h infusion concentration. In pretrained rats, the dose–response function was shifted such that, in watermaze 2, spatial learning was normal at this low concentration, with a deficit at higher infusion concentrations. The induction of long-term potentiation in the dentate gyrus in vivo was blocked at all D-AP5 concentrations. Sensorimotor abnormalities sometimes seen with NMDA receptor antagonists were only apparent at the highest concentration. The implication of this paradoxical dissociation between hippocampal NMDA receptor-dependent plasticity and spatial learning is discussed with reference to two rival hypotheses of the impact of pretraining.