The coordination dynamics of learning were examined in a visuomotor tracking task. Participants produced rhythmic elbow flexion-extension motions to learn a visually defined 90A degrees relative phase tracking pattern with an external sinusoidal signal. There were two visuomotor transformation groups, a correct feedback group and a mirrored feedback group with feedback representing the elbow's motion transformed by 180A degrees. In Experiment 1, the to-be-tracked signal and the participant's motion signal were superimposed within a single window display. In Experiment 2, the to-be-tracked signal and participant's signal were presented in separate windows. Before day 1 practice and 24 h after day 2 practice, participants attempted visually defined 0A degrees, 45A degrees, 90A degrees, 135A degrees, and 180A degrees relative phase tracking patterns either with or without visual feedback of the arm's motion. Before practice, only the 0A degrees and 180A degrees tracking patterns were stable. Practice led to a decrease in phase error toward the required 90A degrees relative phase pattern with a corresponding increase in stability in both the experiments. No effect of visual transformation on performance emerged during practice in the single window task, but did emerge in the two window task. The one window training facilitated transfer to the four unpracticed relative phase patterns, whereas the two window training display only facilitated transfer performance to a single unpracticed relative phase pattern. These findings suggest that the perceptual structure determined the degree of learning and transfer and interacted with the visuomotor transformation. The present findings are discussed with reference to how the visual display constrains the coherency of independent signals with regard to learning and transfer and the role of perceptual discrimination processes linked to transfer.