The rostral medial accessory olive (MAO) of the cat was studied by using an ultrastructural technique combining wheat germ agglutinin-coupled horseradish peroxidase (WGA-HRP) anterograde tracing and postembedding GABA immunocytochemistry. One group of cats received a WGA-HRP injection in the posterior interposed nucleus of the cerebellum and another group received an injection in the nucleus of Darkschewitsch. Based on differences in their morphology three types of GABAergic and three types of nonGABAergic terminals were observed. One type of GABAergic terminal was often GABA/WGA-HRP double-labeled in the cerebellar experiments, and one type of nonGABAergic terminal was often WGA-HRP-labeled in the mesodiencephalic experiments. Following injections of WGA-HRP in the cerebellar nuclei virtually all WGA-HRP-labeled terminals were GABA positive. Quantification of these GABA/WGA-HRP-double-labeled terminals showed that 1) 30% of the GABAergic terminals randomly selected from the entire neuropil were double-labeled, 2) 13% of the GABAergic terminals adjacent to perikarya were double-labeled, and 3) 34% of the GABAergic terminals strategically located next to both of the dendritic elements linked by a gap junction were double-labeled. Statistical analysis of the above data showed that significantly fewer GABAergic terminals adjacent to perikarya were double-labeled (P less than .001) than would be expected from the double-labeled proportion of the randomly selected GABAergic terminals. Following injection of WGA-HRP in the nucleus of Darkschewitsch, all WGA-HRP-labeled terminals were GABA-negative. Quantification of these terminals showed that 1) 26% of the randomly selected nonGABAergic terminals were WGA-HRP labeled, 2) 20% of the nonGABAergic terminals adjacent to perikarya were WGA-HRP labeled, and 3) 23% of the nonGABAergic terminals strategically located next to a gap junction were WGA-HRP labeled. No significant differences were found among these populations. Quantification of terminals of both groups of experiments mentioned above showed that GABAergic terminals composed 1) 38% of the randomly selected terminals, 2) 64% of the terminals apposed to perikarya, and 3) 53% of the terminals strategically located next to gap junctions. Statistical analysis of the above data showed that significantly more GABAergic terminals were located adjacent to perikarya (P less than .001) and strategically next to a gap junction (P less than .05) than would be expected from the random GABAergic innervation. The above findings of the GABAergic, cerebellar, and mesodiencephalic input are discussed with regard to their functional role in the neuronal circuitry of the ros