Arm stiffness is a critical factor underlying stable interactions with the environment. When the hand moves freely through space, a stiff limb would most effectively maintain the hand on the desired path in the face of external perturbations. Conversely, when constrained by a rigid surface, a compliant limb would allow the surface to guide the hand while minimizing variations in contact forces. We aimed to identify the physiological basis of stiffness adaptation for these two classes of movement. Stiffness can be regulated by two mechanisms: co-activation of antagonistic muscles and modulation of reflex gains. We hypothesized that subjects would select high stiffness (high co-activation and/or reflex gains) in free space and high compliance (low co-activation and reflex gains) for constrained movements. We measured EMG and the H-reflex during constrained and unconstrained movement of the wrist. As predicted, subjects co-activated antagonist muscles more when performing the unconstrained movement. Contrary to our hypothesis, however, H-reflex amplitude was higher for the constrained movement despite the a priori preference for lower reflex gains in this situation. In addition, the H-reflex depended on the task and the net force exerted by the limb on the environment, rather than showing a simple dependency on the level of muscle activation. Thus, stiffness seems to increase in free space compared with constrained motion through the use of co-activation while spinal loop gains are adjusted to better regulate the influence of afferences on the ongoing movement. These observations support the hypothesis of movement programming in terms of impedance.