1. Accurate overarm throwing requires precise control of joint rotations so that the ball is released at the appropriate time on the appropriate hand trajectory. Inaccuracy in throws, in turn, must result from errors in the control of joint rotations. But do high and low throws result from disorders in the joint rotations that produce the hand trajectory or in those that cause ball release? Are they due to error at a particular joint or to accumulation of errors across a number of joints? The objective was to answer these questions and thereby to gain insight into the CNS control of joint rotations in a skilled arm movement task. 2. Ten subjects--male, right-handed recreational ball players, all accurate throwers--sat with a fixed trunk and threw tennis balls at a 9 x 9 grid of 6-cm target squares 1.5 or 3 m away. Rotations of five arm segments in three dimensions were measured at 1,000 Hz with the magnetic-field search-coil technique. Hand trajectory (translation) was computed from these rotations. 3. The cause of ball high-low inaccuracy was investigated by determining its relation with hand kinematic parameters that could potentially affect it. No statistically significant relation was found between height of ball impact on the target and height of the hand trajectory. In contrast, statistically significant relations appeared between height of ball impact on the target and both hand trajectory length at ball release (for 8 of 10 subjects) and finger and hand orientation in space at ball release (for all 10 subjects). 4. Three hypotheses were proposed to explain the variable finger and hand orientations in space at ball release, i.e., that they resulted from errors in velocity of rotation at one or more proximal joints (wrist, elbow, shoulder), timing of onset of rotation at one or more proximal joints, or timing of ball release (due to incorrect velocity or timing of onset of finger opening). All three mechanisms could result in inappropriate finger and hand orientations in space at ball release, but the pattern of joint space trajectories would be different in each case. 5. High and low throws did not follow the joint space paths predicted by the first two hypotheses. Instead, as predicted by the third hypothesis, a separation of traces occurred when finger extension was plotted against wrist flexion or against elbow extension, e.g., for a given amplitude of wrist flexion, finger extension was large for the high throws and small for the low throws. 6. In agreement, when all throws were considered, a statistically significant (P < 0.005) relation was found between ball impact height on the target and the amplitude of finger extension, for a fixed amplitude of wrist flexion (10 subjects), and for a fixed amplitude of elbow extension (8 subjects). Only two subjects showed a statistically significant relation between ball impact height and the amplitude of wrist flexion, for a fixed amplitude of elbow extension. 7. The separation of finger extension-wrist flexion traces in joint space for high and low throws was due to a difference in the timing of onset of finger rotation with respect to the wrist rather than to an inappropriate velocity of the finger (or the wrist). 8. As expected, all 10 subjects showed statistically significant relations between the time of onset of finger extension and both finger orientation in space at ball release and height of ball impact on the target. 9. It is concluded that in fast arm-only throws made by male recreational ball players, high and low throws do not primarily result from variability in the height of the hand trajectory or from variability in the amplitude of one or more proximal joints due to errors in the velocity or timing of onset of these joint rotations. Instead, in most cases, they result from inappropriate timing of onset of rotation of the fingers with respect to the rotations of the other joints and thus to inappropriate timing of ball release. These findings emphasize the im