Most manual tasks demand a delicate control of the wrist. Sensory information for this control, e.g. about the position and movement velocity of the hand, is assumed to be primarily provided by muscle spindle afferents. It is known that human muscle spindles in relaxed muscles behave as stretch receptors but it is unclear how they discharge during 'natural' hand movements, since their discharges can also be affected by extrafusal contractions and fusimotor activity. We therefore let subjects perform a centre-out-centre key-pressing task on buttons laid out in a 3 x 3 pattern, a task that allowed unconstrained hand and finger movements and required precise control of the wrist. Microneurography recordings from muscle spindle afferents of the wrist extensor muscles were obtained along with wrist kinematics and electromyographic signals. The discharge rates of afferents were more phase advanced than expected on the length of the radial wrist extensor, which acted as an anti-gravity muscle in the key-pressing task. As such, both acceleration and velocity had significant impacts on the discharge rate of primary afferents, velocity on that of secondary afferents, and length had no impact on either afferent type. The response patterns were different for the two types of muscle spindle afferents from the predominantly eccentrically contracting ulnar wrist extensor: muscle length and velocity had significant impacts on the ensemble response of secondary afferents whereas the primary afferents showed highly variable responses. Accordingly, good predictions of the radial ulnar angular velocity were possible from spindle ensemble responses (R(2) = 0.85) whereas length could be predicted only for phases with lengthening of the ulnar wrist extensor. There are several possible explanations for the unexpectedly large phase advance of spindle afferents in the radial wrist extensor. Given the compliance of tendons, for instance, the phase relationship between the muscle fascicle length and the whole muscle length is conjectured to depend on the load. While additional phase advances are advantageous in motor control, it is concluded that if the central nervous system estimates length or velocity of a muscle from its muscle spindle discharges, this would require additional information about not only the concomitant extrafusal and fusimotor drive but also about the mechanical properties of the load on which the muscle acts.