The purpose of this study was to experimentally investigate the applicability of the equilibrium-point hypothesis to the dynamics of single-joint movements. Subjects were trained to perform relatively slow (movement time 600-1000 ms) or fast (movement time 200-300 ms) single-joint elbow flexion movements against a constant extending torque bias. They were instructed to reproduce the same time pattern of central motor command for a series of movements when the external torque could slowly and unpredictably increase, decrease, or remain constant. For fast movements, the total muscle torque was calculated as a sum of external and inertial components. Analysis of the data allowed reconstruction of the elbow joint compliant characteristics at different times during execution of the learned motor command. "Virtual" trajectories of the movements, representing time-varying changes in a central control parameter, were reconstructed and compared with the "actual" trajectories. For slow movements, the actual trajectories lagged behind the virtual ones. There were no consistent changes in the joint stiffness during slow movements. Similar analysis of experiments without voluntary movements demonstrated a lack of changes in the central parameters, supporting the assumption that the subjects were able to keep the same central motor command in spite of externally imposed unexpected torque perturbations. For the fast movements, the virtual trajectories were N-shaped, and the joint stiffness demonstrated a considerable increase near the middle of the movement. These findings contradict an hypothesis of monotonic joint compliant characteristic translation at a nearly constant rate during such movements.