Intrinsic circuitry within the primary somatosensory cortex of the rat was examined in a combined light and electron microscope study. Corticothalamic projection neurons were retrogradely labeled by applying Phaseolus vulgaris leucoagglutinin (PHA-L) into the ventro-posteromedial thalamic nucleus (VPM). Most labeled neurons were pyramidal cells of layer VI. Postsynaptic targets of recurrent axon collaterals originating from these neurons were assessed in layers IV and V. Single labeled cells, complete with recurrent collaterals, could be isolated in "barrels" in which no anterograde transport had taken place. These findings were confirmed by first eliminating thalamocortical projections from the VPM with kainic acid and then applying PHA-L into the same nucleus. This procedure led to selective retrograde accumulation of tracer in layer VI pyramidal cells. Reconstructed portions of labeled axonal trees reached layer IV, bringing numerous boutons to layers IV, V and VI. The boutons had characteristic drumstick-like shapes. In order to identify postsynaptic targets, 4 sections of axons stemming from 3 neurons were reembedded and serially sectioned for electron microscopy. The ultrastructure of 72 asymmetric synapses, all belonging to identified collaterals, was analysed. Of the 72 terminals, 44 (59.5%) ended on dendritic spines and 30 on shafts of dendrites (40.5%). Perikarya were not among the targets. In a subset of the sample, the nature of the target neurons was examined by postembedding immunohistochemistry for gamma-amino butyric acid (GABA) after staining for PHA-L. A total of 42 labeled terminals was found in layers IV and V; 23 (55%) were located on GABA-negative spines and 19 (45%) on dendritic shafts. Only 6 (32%) of the shafts were GABA-positive. The remaining ones were either clearly GABA-negative, or labeled only at background levels (n = 13; 68%). The results show that most synapses of corticothalamic projection neurons found in layers IV and V terminate on spines and shafts of GABA-negative dendrites. This finding suggests that such recurrent collaterals are involved in both excitatory and inhibitory feedback mechanisms.