The information transmission properties of single, de-efferented primary muscle-spindle afferents from the hind limb of the cat were investigated. The gastrocnemius medialis muscle was stretched randomly while recording spike trains from several muscle-spindle afferents in the dorsal root. Two classes of input stimuli were used: (i) Gaussian noise with band-limited flat spectrum, and (ii) Gaussian noise with a more "naturalistic" 1/f(n) spectrum. The "reconstruction" method was used to calculate a lower bound to the information rate (in bits per second) between the muscle spindles and the spinal cord. Results show that in response to the flat-spectrum input, primary muscle-spindle afferents transfer information mainly about high frequencies, carrying 2.12 bits/spike. In response to naturalistic-spectrum inputs, primary muscle-spindle afferents transfer information about both low and high frequencies, with "spiking efficiency" increasing to 2.67 bits/spike. A simple muscle-spindle simulation model was analyzed with the same method, emphasizing the important part played by the intrafusal fiber mechanical properties in information transmission.