The rehabilitation outcome of two methods of gait retraining, was studied in 40 elderly women who had fractured the neck-of-femur. Twenty control subjects received conventional gait retraining, and 20 experimental subjects participated in a treadmill gait retraining program. Rehabilitation outcome was determined by temporal-distance gait parameters and mobility level (house-bound, limited, or unlimited). Muscle strength and range of movement of the lower limb were measured. Previously reported predictors of poor outcome (age, prefracture mobility, admission from sites other than own home, availability of caring person, type of ambulatory aid, presence or absence of senile dementia, and number of concomitant medical diseases) were recorded. Subjects were assessed on admission to the gait retraining program and at discharge from hospital. A 3 x 2 chi-square analysis on discharge mobility level showed the mobility level of the treadmill group to be significantly (p < .05) higher than the mobility level attained by the control group. Analysis of variance showed on significant differences between the two groups at either assessment time. However, a significant increase in within-group variance from admission to discharge (p < .01) was noted, and this led to the analysis of a subgroup containing six pairs of subjects matched for number of predictors of poor outcome. In the subgroup analysis, the treadmill group was superior to the control group in every parameter measured; significant differences were observed in double-support phase of gait, stance/swing ratio of the affected and unaffected legs, strength of hip flexion of the unaffected leg, strength of hip abductors of affected and unaffected legs, and strength of knee extension of the unaffected leg.