Fifteen normal healthy adult female college students served as subjects in the experiment designed to find out the accuracy of output control in human movement such as jumping. Four types of jumping, eyes-open and eyes-closed Sargent jump and standing broad jump, were performed by each subject. Subjects were asked to jump five different stages of distance graduated in 20% of their maximum performances relying only upon their own subjective estimation. Every stage consisted of two sets of jumping, one set including two trials, so that ten sets were randomly given for one type of jumping. Results were as follows; In general, all subjects successfully graded their output as distance. Standing broad jump was easier to grade than Sargent jump, and eyes-open was easier than eyes-closed. In the eyes-open broad jump, subjectively controlled performance well agreed with the objective scale, but in other jumps, subjective performances always exceeded the objective norms. Reproduction errors were calculated in various ways. Absolute errors were almost constant through five stages regardless of the distance jumped. Hence the relative errors were greater in the lower stages than in higher stages. In particular, algebraic relative error showed that subjects constantly jumped higher in the second trial of one set than in the first trial in the lower stages and vice versa in higher stages. High correlation coefficients were found between the distance jumped and the change of knee angle from still standing position, and between the distance jumped and the time duration in which the force larger than body weight were acting vertically (kicking time), while maximum force applied during kicking showed few significant correlations to the distance jumped. This fact suggests that the distance jumped is controlled through knee position sense which, influencing the time factor in the 'impulse', determines the distance jumped.