Factors determining the release probability were examined using whole-cell patch-clamp recording and minimal stimulation (10 impulses, 50 Hz) of individual glutamatergic synapses, containing single release sites, on pyramidal neurones in the CA1 region of hippocampal slices from neonatal rats. Release probability following the first action potential in the burst (P1) varied among the synapses from 0 to 0.87 (mean +/- s.d.; 0.35 +/- 0.28, n = 52) and the average release during the burst (burst pool) varied from 0.4 to 4.1 events (1.7 +/- 0.85, n = 52). Heterogeneity in P1 did not co-vary with that of the burst pool. By selecting burst trials during which only one release event occurred, the vesicle release probability (Pves) at a release site could be determined. It was found to vary considerably among the synapses, from 0.04 to 0.94 (0.43 +/- 0.28, n = 43). This heterogeneity correlated significantly with that of P1 such that more than half of the variation of P1 could be explained by a variation in Pves. The average number of vesicles directly available for release at the onset of the burst (the pre-primed pool) was estimated as the cumulative release up to that point in the burst where a second release event did not produce higher initial release probability than that found in trials where only one vesicle was released. The average pre-primed pool varied among the synapses from 0.4 to 2.1 (1.03 +/- 0.42, n = 43). It co-varied significantly with that of P1 such that it could explain the remaining variation in P1. The difference between the burst pool and the pre-primed pool suggests the presence of a fast (< 100 ms), activity-dependent priming of vesicles. Some synapses (9/52) did not show any initial release (P1 = 0), but release occurred later during the burst ('low frequency mute synapses'). Their behaviour was explained by an absence of a pre-primed pool.