Previous studies have demonstrated that axonal arbors specific for the four main cortical layers - 2/3, 4, 5, and 6 - develop precisely from the outset using activity-independent cues. In macaque primary visual cortex (V1), layer 2/3 is subdivided into layers named 2/3A, 3B, 4A, and 4B, and layer 4 is subdivided into 4Calpha and 4Cbeta. Individual neurons in V1 of mature macaques have axonal arbors that are highly specific for these sublayers. We have studied the prenatal development of laminar and sublaminar specificity of local circuits in macaque V1. Two-hundred thirty-eight neurons were labeled intracellularly in living brain slices prepared from V1 of five prenatal macaque monkeys aged 100 to 145 d postconception (E100-E145). Axonal and dendritic arbors of labeled neurons were reconstructed to assess their relationships to the cortical layers. We find that developing spiny neurons in layers 2-4B and layer 5 specifically target superficial and deep layers without forming "incorrect" branches in layer 4C. Similarly, layer 6 pyramidal neurons that target layer 4C do not form "incorrect" branches in layer 5. These results indicate that specific projections to the main cortical layers develop with a high degree of selectivity, as in other species. However, the development of sublayer-specific projections was not always precise from the outset. Unlike postnatal animals, axons of some prenatal layer 4Cbeta spiny neurons branch in layer 4B. At similar ages, many pyramidal neurons in the upper half of layer 6 have axonal branches in layer 4Calpha as well as 4Cbeta; these projections are specific for 4Cbeta in more mature animals. Also, there is similar "exuberance" in axonal arbors of other layer 6 cell types. Transient projections were also observed in the subplate and to the white matter for cells from all layers, except 4Cbeta. These observations indicate that at least some sublayer-specific projections emerge by elimination of exuberant axonal branches and suggest that they may use activity-dependent mechanisms to identify "correct" target layers. Such cues could be provided by laminar differences in the patterns of spontaneous prenatal activity in the retino-geniculo-cortical network.