The aim of this study was to determine how context and on-line sensory information are combined to control posture in seated subjects submitted to high-jerk, passive linear accelerations. Subjects were seated with eyes closed on a servo-controlled linear sled. They were asked to relax and received brief accelerations either sideways or in the fore-aft direction. The stimuli had an abrupt onset, comparable to the jerk experienced during a minor car collision. Rotation and translation of the head and body were measured using an Optotrak system. In some of the subjects, surface electromyographic (EMG) responses of selected neck and/or back muscles were recorded simultaneously. For each subject, responses were highly stereotyped from the first trial, and showed little sign of habituation or sensitisation. Comparable results were obtained with sideways and fore-aft accelerations. During each impulse, the head lagged behind the trunk for several tens of milliseconds. The subjects' head movement responses were distributed as a continuum in between two extreme categories. The 'stiff' subjects showed little rotation or translation of the head relative to the trunk for the whole duration of the impulse. In contrast, the 'floppy' subjects showed a large roll or pitch of the head relative to the trunk in the direction opposite to the sled movement. This response appeared as an exaggerated 'inertial' response to the impulse. Surface EMG recordings showed that most of the stiff subjects were not contracting their superficial neck or back muscles. We think they relied on bilateral contractions of their deep, axial musculature to keep the head-neck ensemble in line with the trunk during the movement. About half of the floppy subjects displayed reflex activation of the neck muscles on the side opposite to the direction of acceleration, which occurred before or during the head movement and tended to exaggerate it. The other floppy subjects seemed to rely on only the passive biomechanical properties of their head-neck ensemble to compensate for the perturbation. In our study, proprioception was the sole source of sensory information as long as the head did not move. We therefore presume that the EMG responses and head movements we observed were mainly triggered by the activation of stretch receptors in the hips, trunk and/or neck. The visualisation of an imaginary reference in space during sideways impulses significantly reduced the head roll exhibited by floppy subjects. This suggests that the adoption by the central nervous system of an extrinsic, 'allocentric' frame of reference instead of an intrinsic, 'egocentric' one may be instrumental for the selection of the stiff strategy. The response of floppy subjects appeared to be maladaptive and likely to increase the risk of whiplash injury during motor vehicle accidents. Evolution of postural control may not have taken into account the implications of passive, high-acceleration perturbations affecting seated subjects.