1. The human stretch reflex is known to produce a phase advance in the EMG reflexly evoked by sinusoidal stretching, after allowing for the phase lag introduced by simple conduction. Such phase advance counteracts the tendency to tremor introduced by the combined effect of the conduction delay and the slowness of muscle contraction. The present experiments confirm that the EMG advance cannot be attributed solely to the phase advance introduced by the muscle spindles, and show that a major additional contribution is provided by the dynamic properties of individual motoneurones. 2. The surface EMG was recorded from biceps brachii when two different types of sinusoidally varying mechanical stimuli were applied to its tendon at 2-40 Hz. The first was conventional sinusoidal displacement ('stretch'); the spindle discharge would then have been phase advanced. The second was a series of weak taps at 103 Hz, with their amplitude modulated sinusoidally ('modulated vibration'). The overall spindle discharge should then have been in phase with the modulating signal, since the probability of any individual 1 a fibre responding to a tap would increase with its amplitude. The findings with this new stimulus apply to motoneurone excitation by any rhythmic input, whether generated centrally or peripherally. 3. The sinusoidal variation of the EMG elicited by the modulated vibration still showed a delay-adjusted phase advance, but the value was less than that for simple stretching. At 10 Hz the difference was 70-80 deg. This was taken to be the phase advance introduced by the spindles, very slightly underestimated because of the lags produced by tendon compliance in transmitting sinusoidal stretch to the muscle proper. The adjusted phase advance with modulated vibration was taken to represent that introduced by the reflex centres, undistorted by tendon compliance. At 10 Hz the reflex centres produced about the same amount of phase advance as the muscle spindles. 4. At modulation frequencies above 10 Hz the adjusted central phase advance remained approximately constant. However, when the frequency was reduced to below 6 Hz the central phase advance decreased. The depth of EMG modulation (reflex gain) also fell rapidly, starting from a slightly higher frequency. Thus the central phase advance mechanisms behave like a high-pass filter. 5. A simple model of the motoneurone, incorporating synaptic noise and an after-hyperpolarization, was tested with sinusoidal inputs and gave a phase advance over a wide range of frequencies. The effect was tightly linked to two particular facets of the motor discharge; these were the ratio between the stimulus frequency and the mean firing rate (the 'carrier frequency' of the unit), and the coefficient of variation of the interspike interval distribution. The gain rose to a maximum at the carrier frequency, while the phase advance showed a maximum at 0.8 of the carrier. The more regular the discharge, the greater were these effects. The phase advance might increase to above 90 deg, showing that the motoneurone potentially provides a major contribution to the phase advance of the stretch reflex. Related effects have already been observed in other neuronal models and for the discharge of the muscle spindle, without their significance for the motoneurone being appreciated. In essence, a rhythmically firing neurone is particularly affected by a rhythmic stimulus when the two frequencies approximately coincide. 6. Recording from single human motor units confirmed the role of the 'carrier frequency' in determining the phase advance with sinusoidal inputs. In particular, for both stretching and modulated vibration, the phase advance of the response elicited by a fixed sinusoidal stimulus changed appropriately when the firing rate of the unit varied 'spontaneously' over a long recording period. 7. Thus a combination of modelling and experiment has shown that the motoneurones themselves produce a significant phase advance.