Although many studies have investigated reflexes after stimulation of either cutaneous or proprioceptive afferents, much less is known about responses after more natural perturbations, such as stumbling over an obstacle. In particular, the phase dependency of these responses and their relation to the stumbling behavior has received little attention. Hence response strategies during stumbling reactions after perturbations at different times in the swing phase of gait were studied. While subjects walked on a treadmill, a rigid obstacle unexpectedly obstructed the forward sway of the foot. All subjects showed an "elevating strategy" after early swing perturbations and a "lowering strategy" after late swing perturbations. During the elevating strategy, the foot was directly lifted over the obstacle through extra knee flexion assisted by ipsilateral biceps femoris (iBF) responses and ankle dorsiflexion assisted by tibialis anterior (iTA) responses. Later, large rectus femoris (iRF) activations induced knee extension to place the foot on the treadmill. During the lowering strategy, the foot was quickly placed on the treadmill and was lifted over the obstacle in the subsequent swing. Foot placement was actively controlled by iRF and iBF responses related to knee extension and deceleration of the forward sway. Activations of iTA mostly preceded the main ipsilateral soleus (iSO) responses. For both strategies, four response peaks could be distinguished with latencies of approximately 40 ms (RP1), approximately 75 ms (RP2), approximately 110 ms (RP3), and approximately 160 ms (RP4). The amplitudes of these response peaks depended on the phase in the step cycle. The phase-dependent modulation of the responses could not be accounted for by differences in stimulation or in background activity and therefore is assumed to be premotoneuronal in origin. In mid swing, both the elevating and lowering strategy could occur. For this phase, the responses of the two strategies could be compared in the absence of phase-dependent response modulation. Both strategies had the same initial electromyographic responses till approximately 100 ms (RP1-RP2) after perturbation. The earliest response (RP1) is assumed to be a short-latency stretch reflex evoked by the considerable impact of the collision, whereas the second (RP2) has features reminiscent of cutaneous and proprioceptive responses. Both these responses did not determine the behavioral response strategy. The functionally important response strategies depended on later responses (RP3-RP4). These data suggest that during stumbling reactions, as a first line of defense, the CNS releases a relatively aspecific response, which is followed by an appropriate behavioral response to avoid the obstacle.