The force produced by a skeletal muscle in response to a neural command is a complex function of the muscleÍs length, velocity, and stiffness. Does the brain account for these mechanical properties during task performance? If it does, we hypothesized that humans should learn to control a virtual limb faster with virtual muscles than with a non-physiological virtual force-generator. Sixteen subjects practiced moving a virtual limb from rest to a spatial target 280 times. The virtual limb was controlled with muscle activity recorded from the biceps muscle of each subjectÍs dominant arm using surface electromyography. Muscle contractions were done with the arm in a fixed position. In a muscle dynamics group (n=8), the muscle activity excited a virtual muscle that moved the virtual limb towards the spatial target. In a second force-generator group (n=8) the muscle activity was linearly transformed into a force that moved the virtual limb (no muscle dynamics). Subjects were instructed to move the virtual limb as fast as they could and stop as close to the target as possible, thus, skill was defined by a combined measure of speed and accuracy. In both muscle dynamics and force-generator groups, skill increased significantly with practice; however, there were no differences in the rate of skill improvement between the groups (hypothesis rejected). Additionally, there were no differences in the initial or final skill levels between the groups. This suggests that the brain may not use a detailed representation of mechanical muscle properties during motor planning and performance.