1. In barbiturate-anaesthetized cats, the L7 and S1 dorsal and ventral roots were dissected to isolate functionally single afferents identified as primary endings of soleus muscle spindles, and motor filaments which exerted a fusimotor action on the afferents with limited action on extrafusal muscle. Up to seven filaments, with an action on a given primary ending, could be isolated and each was classified as exerting either a predominantly dynamic or static action.2. Combined stimulation of these filaments, at rates up to 200 impulses/s could maintain afferent firing during muscle shortenings at speeds up to 200 mm/s.3. Fusimotor stimulation could also maintain afferent firing at a target frequency of 100 impulses/s during muscle shortenings up to 200 mm/s. The timing, in relation to the onset of shortening, and the rates of fusimotor stimulation were found to be critical in achieving the target frequency.4. Sinusoidal modulation of the frequency of fusimotor stimulation was used to study the conditions required to achieve constant afferent firing in the face of imposed sinusoidal length changes.5. For given depths of modulation, the phase advance of fusimotor stimulation needed to produce minimum modulation of afferent firing (best compensation) increased with increasing frequency of the sinusoids. The compensation deteriorated with an increase in the frequency of the sinusoids and a change in the mean muscle lengths, although in some cases it could be restored by adjustments to the depth of modulation of fusimotor rate. This suggests that for movements of varying speeds and amplitudes, settings which are appropriate for shortening at a given velocity and mean muscle length, do not apply if either of these two variables are altered.6. These findings demonstrate that the fusimotor system is potentially capable of eliciting constant afferent firing as envisaged in the ;servo-assistance' hypothesis (Matthews, 1964, 1972; Stein, 1974). This, and the fact that constant afferent firing is not seen during normal unobstructed shortenings at velocities greater than 0.2 resting length/s (Prochazka, 1981), are used to argue that it is by choice rather than necessity that ;servo-assistance' (as defined above) is not employed during normal movements. However, servo-assistance of a different form (involving modulated spindle afferent feed-back from both agonists and antagonists) remains a viable alternative.