The representations of the surfaces of the hand in the primary somatosensory cortical field, area 3b, were reconstructed in detail in seven owl monkeys and two squirrel monkeys trained to pick up food pellets from five wells of different sizes. From an early clumsy performance in which several to many retrieval attempts were required for each successful pellet retrieval, the monkeys exhibited a gradual improvement in digital dexterity as shown by significant decreases in mean numbers of grasp attempts/successful retrieval and corresponding standard deviations (e.g. 5.8 +/- 4.5 and 4.8 +/- 3.1 respectively, for the smallest well) between the first and last training sessions. All monkeys commonly used alternative, specific retrieval strategies involving various combinations of digits for significant time epochs before developing a highly successful strategy, which, once achieved, was rapidly stereotyped. For example, the numbers of digit combinations used during the first five versus the last five training sessions decreased from 3.3 +/- 0.7 to 1.8 +/- 0.6 for the smallest well. In both owl and squirrel monkeys, as the behavior came to be stereotyped, monkeys reliably engaged limited surfaces of the glabrous tips of two digits (in eight monkeys), or of three digits (in one monkey) in the palpation and manipulation of these small pellets for their location, capture, and transportation to the mouth. In cortical area 3b, the magnification of representation of these differentially engaged glabrous fingertip surfaces was nearly 2x larger than for the corresponding surfaces of other hand digits, or for the contralateral cortical representations of the same digit surfaces on the opposite hand. In parallel, cutaneous receptive field for area 3b neurons representing crucial digital tip surfaces were less than half as large as were those representing the corresponding surfaces of control digits. Receptive field overlaps were smaller on the trained fingertips than on control fingers. Moreover, the proportion of small overlaps was greater for the trained digits (76 +/- 7%) than for the other digits of the same hand (49 +/- 5.4%). There was still a simple, single--but apparently topologically expanded--representation of these differentially engaged skin surfaces in these monkeys. Thus, with very limited manual exercise over a total period of a few hours of practice at a skill played out in brief daily sessions over a several week long training period, the representations of skin surfaces providing information crucial for successfully performing a small-object retrieval behavior appeared to be substantially remodeled in the most 'primary' of the SI somatosensory cortical fields, cortical area 3b. By that remodeling, behaviorally important skin surfaces were represented in a much finer representational grain than normal. Some implications of these findings for motor skill acquisition are discussed.