1. The triceps surae (TS) stretch reflex was measured in decerebrate cats during crossed extensor stimulation and after spinalization during rhythmic locomotor activity induced by clonidine and manual perineal stimulation. The TS force in response to sinusoidal stretch was measured at a given contraction level before and after deafferentation, and the 'reflex force' was computed by subtracting these two responses. Reflex 'gain' was computed as the ratio of the reflex and deafferented force responses (a unitless estimate of the open loop feedback gain). 2. Prior to locomotion the spontaneous muscle activity was low (less than 15% of maximum), but the reflex gain was relatively high (close to 1.0 with a 5 Hz stretch). When locomotion commenced the reflex gain was markedly lowered when measured at the same contraction level as before locomotion (25% of the gain prior to locomotion). At higher contraction levels the reflex gain was not significantly increased. The reflex force and EMG responses to stretch increased with the contraction level, but their effect on the total reflex gain was cancelled by an associated increase in the intrinsic muscle stiffness. 3. In the decerebrate cat, during weak tonic contractions (spontaneous), the reflex gain was high and comparable with the gain in the resting spinal cat. However, with increased tonic contractions produced by crossed extensor stimulation the reflex gain dropped. At higher contraction levels the gain was not significantly different from the gain during spinal locomotion. 4. When the frequency of stretch was increased from 3 to 20 Hz, EMG responses to stretch increased, but the reflex force decreased, since a more fused contraction developed with the more frequent reflex activations. Overall, the reflex gain decreased with frequency in both spinal and decerebrate cats. The phase lag of the reflex force, relative to the intrinsic muscle force, increased with increasing frequency, due to reflex delays, with a 180 deg lag occurring between 12 and 18 Hz (tremor frequencies). The mean gain was significantly lower and the phase lag was significantly greater during locomotion than during tonic crossed extensor contractions, suggesting different reflex mechanisms. 5. In conclusion, during locomotion in spinal cats afferent feedback from low frequency ankle movements, similar to those occurring during the normal step cycle, reflexly produces a small but significant fraction of the extensor force (about a quarter of the stretch-related force modulation). This fraction is remarkably constant at the different contraction levels of the step cycle. Afferent feedback during higher frequency movement is less effective, minimizing the chance of instability and tremor. In contrast during tonic contractions afferent feedback produces half of the total muscle force during perturbations, clearly contributing to the maintenance of posture.