We confirm Craik's (1947) observation that the human manual1y tracking a visual target behaves like an intermittent servo-control1er. Such tracking responses are indicative of "sampled" negative-feedback control but could be the result of other, continuous, mechanisms. Tracking performance therefore was recorded in a task in which visual feedback of the position of the hand-held joystick could be eliminated. Depriving the subjects of visual feedback led to smoother tracking and greatly reduced the signal power of their responses between 0.5-1.8 Hz. Their responses remained intermittent when they used feedback of their own position but not of the target to track a remembered (virtual) target. Hence, intermittency in tracking behavior is not exclusively a signature of visual feedback control but also may be a sign of feedback to memorized waveforms. Craik's (1947) suggestion that the intermittency is due to a refractory period following each movement was also tested. The errors measured at the start of each intermittent response, during tracking of slow waveforms, showed evidence of a small error deadzone (measuring 0.7 cm on the VDU screen or 0.80 degrees at the eye). At higher target speeds, however, the mean size of starting errors increased, and the upper boundary of the distribution of starting error was close to that expected of a refractory delay of approximately 170 ms between responses. We consider a model of the control system that can fit these results by incorporating an error deadzone within a feedback control loop. We therefore propose that the initiation of intermittent tracking responses may be limited by a positional error deadzone and that evidence for a refractory period between successive corrective movements can be satisfied without evoking an explicit timing or sampling mechanism.