1. Long-term potentiation (LTP) of synaptic transmission is the putative mechanism underlying learning and memory. Despite intensive study, it remains controversial whether LTP is expressed at a pre- or postsynaptic locus. A new approach was used to investigate this question at excitatory synapses from the medial perforant path (MPP) onto granule cells in the hippocampal dentate gyrus. The variance of the evoked synaptic amplitude was plotted against mean synaptic amplitude at several different Cd2+ concentrations. The slope of the variance-mean plot estimates the average amplitude of the response following the release of a single vesicle of transmitter (Qav). A presynaptic modulation should not affect Qav, but a postsynaptic modulation should alter it. 2. The variance-mean technique was tested by applying the analysis before and after three different synaptic modulations: (i) a reduction in Qav by the addition of the competitive antagonist CNQX; (ii) a reduction in the average probability of transmitter release (Pav) by the addition of baclofen; and (iii) an increase in the number of active synaptic terminals (N) by increasing the stimulus strength. CNQX reduced the average synaptic amplitude and Qav to the same extent, consistent with a postsynaptic action. In contrast, neither a change in N nor Pav altered Qav. This confirms that the variance-mean technique can distinguish between a pre- and a postsynaptic site of modulation. 3. Induction of LTP increased EPSC amplitude by 50 +/- 0.4 % (n = 5) and, in the same cells, increased Qav by 47 +/- 0.6 %. There was no significant difference between the increase in EPSC amplitude and the increase in Qav. Thus, LTP of the MPP input to dentate granule cells can be explained by an increase in the postsynaptic response to transmitter.