The role of reflexes in the control of stiffness during human elbow joint movement was investigated for a wide range of movement speeds (1.5-6 rad/s). The electromyographic (EMG) responses of the elbow joint muscles to step position errors (step amplitude 0.15 rad; rise time 100 ms) imposed at the onset of targeted flexion movements (1.0 rad amplitude) were recorded. For all speeds of movement, the step position disturbance produced large modulations of the usual triphasic EMG activity, both excitatory and inhibitory, with an onset latency of 25 ms. In the muscles stretched by the perturbation, the early EMG response (25-60 ms latency) magnitude was greater than 50% of the activity during the unperturbed movements (background activity). In all muscles the EMG responses integrated over the entire movement were greater than 25% of the background activity. The responses were relatively greater for slower movements. Perturbations assisting the movement caused a short-latency (25-60 ms) reflex response (in the antagonist muscle) that increased with movement speed and was constant as a percentage of the background EMG activity. In contrast, perturbations resisting the movement caused a reflex response (in the agonist muscle) that was of the same absolute magnitude at all movement speeds, and thus decreased with movement speed as a percentage of the background EMG activity. There was a directional asymmetry in the reflex response, which produced an asymmetry in the mechanical response during slow movements. When the step perturbation occurred in a direction assisting the flexion movement, the antagonist muscle activity increased, but the main component of this response was delayed until the normal time of onset of the antagonist burst. When the step perturbation resisted the movement the agonist muscles responded briskly at short latency (25 ms). A reflex reversal occurred in two of six subjects. A fixed reflex response occurred in the antagonist muscle, regardless of the perturbation direction. For the extension direction perturbations (resisting movement), this response represented a reflex reversal (50 ms onset latency) and it caused the torque resisting the imposed step (stiffness) to drop markedly (below zero for one subject). Reflex responses were larger when the subject was prevented from reaching the target. That is, when the perturbation remained on until after the normal time of reaching the target, the EMG activity increased, with a parallel increase in stiffness. Similarly, when the perturbations prevented the subject from reaching the target during a 1-rad voluntary cyclic movement, the EMG and stiffness increased markedly.(ABSTRACT TRUNCATED AT 400 WORDS)