An important theoretical concept in motor control is the idea that the CNS uses an internal model of the motor system and environment to predict the sensory consequences of motor commands. In arm movement control, a critical factor affecting the transformation from motor commands to sensory consequences is limb dynamics, including the inertial anisotropy of the arm, which refers to the fact that the inertial resistance of the arm depends on hand movement direction. Here we show that the CNS maintains an accurate internal model of the inertial anisotropy of the arm by demonstrating that the motor system can precisely predict direction-dependent variations in hand acceleration. Subjects slid an object, held beneath the index finger, across a frictionless horizontal surface to radially located targets. We recorded the normal (vertical) force exerted by the fingertip, as well as the tangential (horizontal) force proportional to hand acceleration. We found that normal force was precisely scaled in anticipation of tangential force, which, as expected, varied with direction. The peak rates of change of the normal and tangential forces, observed early in the movement, were highly correlated. Similar results were obtained regardless of whether the start position of the hand was located directly in front of the subject or rotated 45 degrees to the right. Finally, we observed reduced force correlations under reaction time conditions. This suggests that the process of prediction, based on an internal model of the limb, is not fully completed within the reaction time interval.