In order to understand how the central nervous system controls the kinematics of rapid finger and hand movements, we studied the motions of subjects turning a knob to light-emitting diode targets, similar to tuning a radio dial. On many trials, subjects turned the knob with a single, smooth, and regular motion as revealed by the angular position and velocity trajectories, but on others, subjects produced irregularities in the kinematics. Like many past studies, we interpreted these irregularities as discrete corrective submovements. Unlike other studies, we used a direct, objective algorithm to identify overlapping submovements, detecting appreciable inflections in the acceleration traces by examining zero crossings in their derivatives, jerk and snap. The movements without overlapping submovements on average had a near symmetric, bell-shaped velocity profile that was independent of speed, and which matched the theoretical minimum jerk velocity very closely. We proposed three plausible mechanisms for altering the shape of movement kinematics, and implemented a mass-spring model with nonlinear damping to explore the possibilities. Although there was relatively little variability in the shape and symmetry of movements across trials, there was a fair amount of variability in their amplitude. We show that subjects attempted to eliminate the need for corrective submovements by making more accurate primary movements with practice, but that the variability inherent in rapid movements dictated the need for corrective submovements. Subjects used corrective submovements to improve final endpoint accuracy while reducing endpoint variability, resulting in higher task success rates.