In order to study postural maintenance we have used a substantial inverted pendulum with one degree of freedom as an unstable load. In our previous work, such a human-sized inverted pendulum has been balanced pedally (Loram et al. 2001) or manually (Lakie et al. 2003). The advantage of the manual approach is that it allows the movement not only of the load but also of the actuator (the hand) to be directly recorded and the stiffness of the coupling between actuator and load to be altered. The inverted pendulum represents the body, the hand represents the calf muscles, and the spring coupling the two represents the stiffness of the Achilles tendon and foot. It is a reductionist approach where the role of the human operator becomes solely the transformation of the available sensory data into appropriate motor acts. Because the head is stationary the sensory information that is available to the subject can be precisely controlled. Although it is not ‘real’ standing, dynamic ultrasonography (Loram et al. 2005) has revealed the close similarities between movements of the hand in manual balance and movements of the calf muscles in naturally standing subjects. In balancing an inverted pendulum, and in real standing, the active postural muscles exhibit repeated impulsive surges of activity which change the length (bias) of the series elastic component (s.e.c.). These purposive, bias adjustments occur more frequently than body sway and they must be pre-planned by the nervous system which has available recent sensory information concerning sway (Lakie et al. 2003; Loram et al. 2005). It is well known that impoverished sensory information leads to impaired control of balance. How does the frequency of these bias adjustments relate to the quantity of sensory information available? Craik (1947) was the first to investigate this question using an entirely different task which was visually controlled manual tracking of a moving target. He reasoned that if corrective manual adjustments depended on the visual detection of a disparity then an increase in visual gain would cause error to accrue more rapidly with a consequent increase in the frequency of adjustments. In fact, he found that the frequency of manual adjustments was substantially independent of visual gain. Accordingly, he concluded that a central movement planning process rather than the availability of sensory data limited the frequency of adjustments. Here we investigate the ‘Craik’ question in a balancing task. We systematically reduced the availability of sensory information from the visual, vestibular and somatosensory senses. If it is necessary that sensory information must attain a threshold value (or a critical signal/noise ratio) before a balancing impulse can be generated then a reduction in sensory information should reduce the frequency of bias adjustments. Ten subjects performed simple tasks analogous to human standing with specific senses operating. In every case balancing was controlled by a spring of inadequate stiffness for passive stability (nominally 85% of the load stiffness). This stiffness simulates the stiffness of the s.e.c. which we believe is limited by the tendon and foot. Subjects balanced by manually altering the length of the spring (bias adjustments). The sensory information that was available to them could be purely visual, purely vestibular, purely proprioceptive or a combination of two or more of these. The movement of the body or load (sway) and the bias adjustments were recorded. Subsequent analysis showed that the mean frequency of bias adjustments (~2.4 s^-1) decreased trivially as sensory information was limited. Therefore the mean duration (~400 ms) of bias adjustments is not determined by the need to reach a sensory threshold. Reducing sensory information always increased mean sway size and this is presumably due to poorer judgement in the planning of each bias adjustment. In additional experiments, measurements of bias adjustments in the triceps surae were made in standing subjects with and without vision. The mean bias adjustment frequency was 2.6 Hz in both cases. Manual balancing and real standing appear to operate on similar principles. We suggest that the frequency of bias adjustments is related to a central planning time which is common to vision, vestibular and proprioceptive senses and has much in common with intermittency in the voluntary control of constant force (Slifkin et al. 2000) as well as in manual tracking of a visual target. Sway of the inverted pendulum or postural sway of the body may be a consequence of purposive centrally controlled repetitive adjustments in muscle output which act to regulate position and velocity and thus underlie the maintenance of posture.