1. Dual intracellular recordings were made from synaptically coupled pyramidal cell-to-interneurone pairs (n = 5) of the cat visual cortex in vitro. Pre- and postsynaptic neurones were labelled with biocytin, followed by correlated light and electron microscopic analysis to determine all sites of synaptic interaction. 2. Pyramidal neurones in layers II-III elicited monosynaptic EPSPs in three distinct classes of smooth dendritic local-circuit neurones, namely basket cells (n = 3), a dendrite-targeting cell (n = 1) and a double bouquet cell (n = 1). Unitary EPSPs in basket cells were mediated by one, two, and two synaptic junctions, whereas the pyramid-to-dendrite-targeting cell and pyramid-to-double bouquet cell interaction were mediated by five and seven synaptic junctions, respectively. Recurrent synaptic junctions were found on all somato-dendritic compartments, with a tendency to be clustered close to the soma on the double bouquet and dendrite-targeting cells. The latter interneurones were reciprocally connected with pyramidal cells. 3. Unitary EPSPs had an average peak amplitude of 1005 +/- 518 microV, fast rise times (10-90%; 0.67 +/- 0.25 ms) and were of short duration (at half-amplitude, 4.7 +/- 1.0 ms). Their decay was monoexponential (tau = 7.8 +/- 4.3 ms) at hyperpolarized membrane potentials and appeared to be shaped by passive membrane properties (tau = 9.2 +/- 8.5 ms). All parameters of concomitantly recorded spontaneous EPSPs were remarkably similar (mean amplitude, 981 +/- 433 microV; mean rise time, 0.68 +/- 0.18 ms; mean duration, 4.7 +/- 1.7 ms). 4. In all three pyramidal-to-basket cell pairs, closely timed (10-50 ms) pairs of presynaptic action potentials resulted in statistically significant paired-pulse depression, the mean of the averaged second EPSPs being 80 +/- 11% of the averaged conditioning event. The overall degree of paired-pulse modulation was relatively little affected by either the amplitude of the preceding event or the inter-event interval. 5. The probability density function of the peak amplitudes of the unitary EPSPs could be adequately fitted with a quantal model. Without quantal variance, however, the minimum number of components in the model, excluding the failures, exceeded the number of electron microscopically determined synaptic junctions for all five connections. In contrast, incorporating quantal variance gave a minimum number of components which was compatible with the number of synaptic junctions, and which fitted the data equally well as models incorporating additional components but no quantal variance. For this model with quantal variance with the minimum number of components the estimate of the quantal coefficient of variation ranged between 0.33 and 0.46, and the corresponding quantal sizes ranged between 260 and 657 microV. The peak EPSP amplitudes in two of the four connections with more than one synaptic junction could be adequately described by a uniform binomial model for transmitter release. 6. In conclusion, at least three distinct interneurone classes receive local excitatory pyramidal cell input which they relay to different compartments on their postsynaptic target neurones. The reliability of transmission is high, but the fast time course of the EPSPs constrains their temporal summation. Due to the relatively small amplitude of unitary EPSPs several convergent inputs will therefore be required to elicit suprathreshold responses.