1. The effects of synapses established by smooth dendritic neurones on pyramidal and spiny stellate cells were studied in areas 17 and 18 of the cat visual cortex in vitro. Paired intracellular recordings with biocytin-filled electrodes and subsequent light and electron microscopic analysis were used to determine the sites of synaptic interaction. 2. All smooth dendritic cells established type II synapses previously shown to be made by terminals containing GABA, therefore the studied cells are probably GABAergic. Three classes of presynaptic cell could be defined, based on their efferent synaptic target preference determined from random samples of unlabelled postsynaptic cells. (a) Basket cells (n = 6) innervated mainly somata (49.9 +/- 13.8%) and dendritic shafts (45.2 +/- 10.7%) and, to a lesser extent, dendritic spines (4.9 +/- 4.6%). (b) Dendrite-targeting cells (n = 5) established synapses predominantly on dendritic shafts (84.3 +/- 9.4%) and less frequently on dendritic spines (11.2 +/- 6.7%) or somata (4.5 +/- 4.7%). (c) Double bouquet cells (n = 4) preferred dendritic spines (69.2 +/- 4.2%) to dendritic shafts (30.8 +/- 4.2%) as postsynaptic targets and avoided somata. 3. Interneurones formed 5240 +/- 1600 (range, 2830-9690) synaptic junctions in the slices. Based on the density of synapses made by single interneurones and the volume density of GABAergic synapses, it was calculated that an average interneurone provides 0.66 +/- 0.20% of the GABAergic synapses in its axonal field. 4. The location of synaptic junctions on individual, identified postsynaptic cells reflected the overall postsynaptic target distribution of the same GABAergic neurone. The number of synaptic junctions between pairs of neurones could not be predicted from light microscopic examination. The number of electron microscopically verified synaptic sites was generally smaller for the dendritic domain and larger for the somatic domain than expected from light microscopy. All presynaptic cells established multiple synaptic junctions on their postsynaptic target cells. A basket cell innervated a pyramidal cell via fifteen release sites; the numbers of synapses formed by three dendrite-targeting cells on pyramidal cells were seventeen and eight respectively, and three on a spiny stellate cell; the interaction between a double bouquet cell and a postsynaptic pyramidal cell was mediated by ten synaptic junctions. 5. All three types of interneurone (n = 6; 2 for each type of cell) elicited short-latency IPSPs with fast rise time (10-90%; 2.59 +/- 1.02 ms) and short duration (at half-amplitude, 15.82 +/- 5.24 ms), similar to those mediated by GABAA receptors. 6. Average amplitudes of unitary IPSPs (n = 6) were 845 +/- 796 microV (range, 134-2265 microV). Variability of IPSP amplitude was moderate, the average ratio of IPSP and baseline noise variance was 1.54 +/- 0.96. High frequency activation of single presynaptic dendrite-targeting cells led to an initial summation followed by use-dependent depression of the averaged postsynaptic response. Double bouquet cell-evoked IPSPs, recorded in the soma, had a smaller amplitude than those evoked by the other two cell types. In all connections, transmission failures were rare or absent, particularly when mediated by a high number of release sites. 7. The results demonstrate that different types of neocortical GABAergic neurones innervate distinct domains on the surface of their postsynaptic target cells. Nevertheless, all three types of cell studied here elicit fast IPSPs and provide GABAergic input through multiple synaptic release sites with few, if any, failures of transmission.