Corticospinal projections in adult rodents arise exclusively from layer V neurons in the sensorimotor cortex. These neurons are topographically organized in their connections to spinal cord targets. Previous studies in rodents have shown that the mature distribution pattern of corticospinal neurons develops during the first 2 weeks postnatal from an initial widespread pattern that includes the visual cortex to a distribution restricted to the sensorimotor cortex. To determine whether specificity in corticospinal connections also emerges from an initially diffuse set of projections, we have studied the outgrowth of corticospinal axons and the formation of terminal arbors in developing hamsters. The sensitive fluorescent tracer 1,1',dioctadecyl-3,3,3',3'-tetramethylindocarbocyanine perchlorate (DiI) was used to label corticospinal axons from the visual cortex or from small regions of the forelimb or hindlimb sensorimotor cortex in living animals at 4-17 days postnatal. Initially axon outgrowth was imprecise. Some visual cortical axons extended transiently beyond their permanent targets in the pontine nuclei, by growing through the pyramidal decussation and in some cases extending as far caudally as the lumbar enlargement. Forelimb sensorimotor axons also extended past their targets in the cervical enlargement, in many cases growing in the corticospinal tract to lumbar levels of the cord. By about 17 days postnatal these misdirected axons or axon segments were withdrawn from the tract. Despite these errors in axon trajectories within the corticospinal tract, terminal arbors branching into targets in the spinal gray matter were topographically appropriate from the earliest stages of innervation. Thus visual cortical axons never formed connections in the spinal cord, forelimb sensorimotor axons arborized only in the cervical enlargement, and hindlimb cortical axons terminated only in the lumbar cord at all stages of development examined. Corticospinal arbors formed from collaterals that extended at right angles from the shafts of primary axons, most likely by the process of interstitial branching after the primary growth cone had extended past the target. Once collaterals extended into the spinal gray matter, highly branched terminal arbors formed within 2-4 days, beginning at about 4 and 8 days postnatal for the cervical and lumbar enlargements, respectively. These results show that specificity in corticospinal connectivity is achieved by selective growth of axon collaterals into appropriate spinal targets from the beginning and not by the later remodeling of initially diffuse connections. In contrast, errors occur in the initial outgrowth of axons in the corticospinal tract, which are subsequently corrected.