We studied the sources of thalamic projections to prefrontal areas of nine rhesus monkeys with the aid of retrograde tracers (horseradish peroxidase or fluorescent dyes). Our goal was to determine the proportion of labeled neurons contributing to this projection system by the mediodorsal (MD) nucleus compared to those distributed in other thalamic nuclei, and to investigate the relationship of thalamic projections to specific architectonic areas of the prefrontal cortex. We selected areas for study within both the basoventral (areas 11, 12, and ventral 46) and the mediodorsal (areas 32, 14, 46, and 8) prefrontal sectors. This choice was based on our previous studies, which indicate differences in cortical projections to these two distinct architectonic sectors (Barbas, '88; Barbas and Pandya, '89). In addition, for each sector we included areas with different architectonic profiles, which is also relevant to the connectional patterns of the prefrontal cortices. The results showed that MD included a clear majority (over 80%) of all thalamic neurons directed to some prefrontal cortices (areas 11, 46, and 8); it contributed just over half to some others (areas 12 and 32), and less than a third to area 14. Clusters of neurons directed to basoventral and mediodorsal prefrontal areas were largely segregated within MD: the former were found ventrally, the latter dorsally. However, the most striking findings establish a relationship between thalamic origin and laminar definition of the prefrontal target areas. Most thalamic neurons directed to lateral prefrontal cortices, which are characterized by a high degree of laminar definition (areas 46 and 8), originated in the parvicellular and multiform subdivisions of MD, and only a few were found in other nuclei. In contrast, orbital and medial cortices, which have a low degree of laminar differentiation, were targeted by the magnocellular subdivision of MD and by numerous other limbic thalamic nuclei, including the midline and the anterior. Thus topographic specificity in the origin of thalamic projections increased as the laminar definition of the target area increased. Moreover, the rostrocaudal distribution of labeled neurons in MD and the medial pulvinar also differed depending on the degree of the laminar definition of the prefrontal target areas. The rostral parts of MD and the medial pulvinar projected to the eulaminate lateral prefrontal cortices, whereas their caudal parts projected to orbital and medial limbic cortices. Selective destruction of caudal MD is known to disrupt mnemonic processes in both humans and monkeys, suggesting that this thalamic-limbic prefrontal loop may constitute an important pathway for memory.