It is not yet certain which sources of information are most important in judging the weight of a held object. In order to study this question further, a "deafferented" man and five controls flexed their wrist to lift a container weighing 1000 g. Direct vision of the arm and weight was denied; the container's vertical position was displayed to the subjects on an oscilloscope at the start of each trial and, then, in most experimental conditions, this display was removed. The weight was then either gradually increased or decreased over 20 s or left unchanged, on a pseudorandom basis. A verbal judgement of its change was required at the end of each trial, lasting 20 or 40 s. Under these conditions, the "deafferented" subject was unable to correctly judge the weight changes (38% accuracy, n.s. χ2, compared with 77% in control subjects), and even the control subjects, when exposed to muscle vibration, made many errors (54% accuracy). However, in many trials, including those in which the weight was unchanged, the vertical height of the container was not held constant by the subjects, but drifted up or down (mean absolute drift: approximately 2 cm). Hence, the change in muscular activation or stiffness could be estimated by the observers in the majority of trials. This allowed the verbal judgements of both the "deafferented" man and of control subjects undergoing muscle vibration to be correlated with the muscle activation produced, independent of the actual weight being tested. Post-hoc predictions of controls' responses during vibration, based on the direction of the change in muscle activity which these drifts in position implied, were 77% and 66% accurate for ±750 g and ±375 g tasks and 73% accurate for forearm-vibration trials (P<0.0001, χ2). Predictions of the "deafferented" subject's responses were 64% accurate (P=0.0002, χ2), even though his own responses were at a chance level with respect to the actual weight change. The judgements made by these subjects might have been based upon a peripheral sensory input, as small afferent fibres are still present in the "deafferented" man and vibration only partly blocked sensory function in the control subjects. Care was taken to minimise all other possible cues to the weight changes, e.g. vestibular, thermal, pressure or pain cues. However, peripheral inputs may not be the only signals used in the subjects' perceptual judgements. They might, instead, be based upon a centrally originating, but illusory changing sense of body position or, possibly, a changing sense of effort. In both cases, a perceived discordance between voluntary muscle activation and body image could underlie the subjects' responses. Our data do not yet allow us to distinguish between these alternative peripheral and central hypotheses, but do highlight the need to include perceptions of body position and motion into judgements of force control.