It is widely believed that synaptic plasticity may provide the neural mechanism that underlies certain kinds of learning and memory in the mammalian brain. The expression of long-term potentiation (LTP) in the hippocampus, an experimental model of synaptic plasticity, requires the GluR-A subunit of the AMPA subtype of glutamate receptor. Genetically modified mice lacking the GluR-A subunit show normal acquisition of the standard, fixed-location, hidden-platform watermaze task, a spatial reference memory task that requires the hippocampus. In contrast, these mice are dramatically impaired on hippocampus-dependent, spatial working memory tasks, in which the spatial response of the animal is dependent on information in short-term memory. Taken together, these results argue for two distinct and independent spatial information processing mechanisms: (i) a GluR-A-independent associative learning mechanism through which a particular spatial response is gradually or incrementally strengthened, and which presumably underlies the acquisition of the classic watermaze paradigm and (ii) a GluR-A-dependent, non-associative, short-term memory trace which determines performance on spatial working memory tasks. These results are discussed in terms of Wagner's SOP model (1981).