Aim To examine the efficacy of postural strategy training using a balance exercise assist robot (BEAR) as compared with conventional balance training for frail older adults. Methods The present study was designed as a cross‐over trial without a washout term. A total of 27 community‐dwelling frail or prefrail elderly residents (7 men, 20 women; age range 65–85 years) were selected from a volunteer sample. Two exercises were prepared for interventions: robotic exercise moving the center of gravity by the balance exercise assist robot system; and conventional balance training combining muscle‐strengthening exercise, postural strategy training and applied motion exercise. Each exercise was carried out twice a week for 6 weeks. Participants were allocated randomly to either the robotic exercise first group or the conventional balance exercise first group. Main outcome measures: preferred and maximal gait speeds, tandem gait speeds, timed up‐and‐go test, functional reach test, functional base of support, center of pressure, and muscle strength of the lower extremities were assessed before and after completion of each exercise program. Results Robotic exercise achieved significant improvements for tandem gait speed (P = 0.012), functional reach test (P = 0.002), timed up‐and‐go test (P = 0.023) and muscle strength of the lower extremities (P = 0.001–0.030) compared with conventional exercise. Conclusions In frail or prefrail older adults, robotic exercise was more effective for improving dynamic balance and lower extremity muscle strength than conventional exercise. These findings suggest that postural strategy training with the balance exercise assist robot is effective to improve the gait instability and muscle weakness often seen in frail older adults. Geriatr Gerontol Int 2017; ••: ••–••.