For sensorimotor transformations to be executed accurately, there must be mechanisms that can both establish and modify mappings between sensory and motor coordinates. Such mechanisms were investigated in normal subjects using a reach-to-grasp task. First, we replaced the normal input of visual information about object size with auditory information, i.e., we attempted to establish an ‘audiomotor map’. The size of the object was log linearly related to the frequency of the sound, and we measured the maximum grip aperture (MGA) during the reaching phase to determine if the subjects had learned the relationship. Second, we changed the frequency-object size relationship to study adaptation in the newly acquired map. Our results demonstrate that learning of an audiomotor map consisted of three distinct phases: during the first stage (˜10-15 trials) subjects simply used MGAs large enough to grasp any reasonably sized object and there were no overt signs of learning. During the second stage, there was a period of fast learning where the slope of the relationship between MGA and object size became steeper until the third stage where the slope was constant. In contrast, when sensorimotor adaptation was studied in the established audiomotor map, there was rapid learning from the start of a size perturbation. We conclude that different learning strategies are employed when sensorimotor transformations are established compared to when existing transformations are modified.