The control of pointing arm movements in the absence of visual guidance was investigated in unpracticed human subjects. The right arm grasped a lever which restricted the movement of the right index fingertip to a horizontal arc, centered between the axes of eye rotation. A horizontal panel directly above the arm prevented visual feedback of the movement. Visual stimuli were presented in discrete positions just above panel and fingertip. A flag provided visual feedback on fingertip position before each pointing movement (Exp. A and B), or before a movement sequence (Exp. C). When subjects pointed from straight ahead to eccentric stimulus positions (Exp. A), systematic and variable pointing errors were observed; both kinds of errors increased with stimulus eccentricity. When subjects pointed from 30 deg left to stimuli located further right (Exp. B), errors increased with stimulus position to the right. Taken together, these findings suggest that pointing accuracy depends not primarily on stimulus position, but rather on required movement amplitude. When subjects performed sequences of unidirectional movements (Exp. C), systematic and variable errors increased within the sequence. A quantitative analysis revealed that this increase can be best described as an accumulation of successive pointing errors. We conclude that both findings, error increase with amplitude, and accumulation of successive errors, when considered together strongly support the hypothesis that amplitude, rather than final position, is the controlled variable of the investigated movements.