We investigated the physiological basis of the 'broken escalator phenomenon', namely the sensation that when walking onto an escalator which is stationary one experiences an odd sensation of imbalance, despite full awareness that the escalator is not going to move. The experimental moving surface was provided by a linear motor-powered sled, moving at 1.2 m/s. Sled velocity, trunk position, trunk angular velocity, EMG of the ankle flexors-extensors and foot-contact signals were recorded in 14 normal subjects. The experiments involved, initially, walking onto the stationary sled (condition Before). Then, subjects walked 20 times onto the moving sled (condition Moving), and it was noted that they increased their walking velocity from a baseline of 0.60 m/s to 0.90 m/s. After the moving trials, subjects were unequivocally warned that the platform would no longer move and asked to walk onto the stationary sled again (condition After). It was found that, despite this warning, subjects walked onto the stationary platform inappropriately fast (0.71 m/s), experienced a large overshoot of the trunk and displayed increased leg electromyographic (EMG) activity. Subjects were surprised by their own behaviour and subjectively reported that the 'broken escalator phenomenon', as experienced in urban life, felt similar to the experiment. By the second trial, most movement parameters had returned to baseline values. The findings represent a motor aftereffect of walking onto a moving platform that occurs despite full knowledge of the changing context. As such, it demonstrates dissociation between the declarative and procedural systems in the CNS. Since gait velocity was raised before foot-sled contact, the findings are at least partly explained by open-loop, predictive behaviour. A cautious strategy of limb stiffness was not responsible for the aftereffect, as revealed by no increase in muscle cocontraction. The observed aftereffect is unlike others previously reported in the literature, which occur only after prolonged continuous exposure to a sensory mismatch, large numbers of learning trials or unpredictable catch trials. The relative ease with which the aftereffect was induced suggests that locomotor adaptation may be more impervious to cognitive control than other types of motor learning.