Maintenance of upright, human balance is neurologically and biomechanically a complex process, though the ankle strategy predominates in quiet standing. This investigation seeks insight into the complex problem by studying a reduced, yet related problem of how the ankle mechanisms are used to balance a human proportioned inverted pendulum. A distinguishing feature of the task is that despite one's best efforts to control this unstable load some irreducible sway always remains. Contrary to published ideas, modulation of effective ankle stiffness was not the way that sway size was altered. Rather, position was controlled by an intermittent, neurally modulated, ballistic-like pattern of torque whose anticipatory accuracy was improved to reduce sway size. Using a model, and by direct measurement, I found the intrinsic mechanical ankle stiffness will only partially counter the "gravitational spring". Since this stiffness was substantially constant and cannot be neurally modulated, I attribute it to the foot, tendon and aponeurosis rather than the activated calf muscle fibres. Thus triceps-surae muscles maintain balance via a spring-like element which is itself generally too compliant to provide even minimal stability. I hypothesise that balance is maintained by anticipatory, ballistic-like, biasing of the series-elastic element resulting from intermittent modulation of the triceps-surae.