The main objective of this study was to characterize the stretch reflex response of the human thigh muscles to an unexpected knee flexion at the transition from stance to swing during walking. Eleven healthy subjects walked on a treadmill at their preferred speed. Reliable and constant knee flexions (6-12 degrees amplitude, 230-350 degrees /s velocity, 220 ms duration) were applied during the late swing and early stance phase of human walking by rotating the knee joint with a specifically designed portable stretch apparatus affixed to the left knee. Responses from rectus femoris (RF), vastus lateralis (VL), vastus medialis (VM), biceps femoris (BF), medial hamstrings (MH) and medial gastrocnemius (GM) were recorded via bipolar surface electromyograms (EMG). The onset of the response in the RF, VL and VM, remained stable and independent of the time in the step cycle when the stretch was applied. Across all subjects the response onset (mean +/- SD) occurred at 23+/-1, 24+/-1 and 23+/-1 ms for RF, VL and VM, respectively. The duration of the initial response was 90-110 ms, at which time the EMG signal returned towards baseline levels. Three reflex response windows, labelled the short latency reflex (SLR), the medium latency reflex (MLR) and the late latency reflex response (LLR), were analysed. The medium and late reflex responses of all knee extensors increased significantly ( p=0.008) as the gait cycle progressed from swing to stance. This was not related to the background EMG activity. In contrast, during standing at extensor EMG levels similar to those attained during walking the reflex responses were dependent on background EMG. During walking, LLR amplitudes expressed as a function of the background activity were on average two to three times greater than SLR and MLR reflex amplitudes. Distinct differences in SLR and LLR amplitude were observed for RF, VL and VM but not in the MLR amplitude. This may be related to the different pathways mediating the SLR, MLR and LLR components of the stretch response. As for the knee extensor antagonists, they exhibited a response to the stretch of the quadriceps at latencies short enough to be monosynaptic. This is in agreement with the suggestion by Eccles and Lundberg (1958) that there may be functional excitatory connections between the knee extensors and flexors in mammals.