In a prospective study of 116 elderly people admitted to a geriatric rehabilitation ward, mini-mental state score on admission was used to classify nonaphasic subjects into three groups: cognitively normal (24-30), mildly impaired (17-23) and severely impaired (less than 17). Forty-three percent of the subjects were normal, 34% had mild and 23% had severe cognitive impairment. Using walking speed as the index of gait rehabilitation, subjects with normal cognition had significantly greater gait improvement than those with cognitive impairment. Age and urinary incontinence did not contribute to predicting gait improvement in regression modelling. Cognitive impairment also predicted discharge placement to long term hospital care and to a higher level of care than the subject received prior to admission. Thus cognitive impairment on admission to a geriatric rehabilitation ward predicts poor rehabilitation outcome with respect to gait and discharge residence.