OBJECTIVE To evaluate the effects of balance retraining in a sample of people with multiple sclerosis. METHODS Randomized controlled trial. METHODS Rehabilitation unit. METHODS A consecutive sample of 44 subjects was randomized into two experimental groups and one control group. The inclusion criteria were: ability to stand independently more than 30 seconds, ability to walk for 6 m. METHODS Group 1 received balance rehabilitation to improve motor and sensory strategies. Group 2 received balance rehabilitation to improve motor strategy. Group 3 received treatments not specifically aimed at improving balance. METHODS Berg Balance Scale, Dynamic Gait Index and fall frequency were used to assess balance impairments. Dizziness Handicap Inventory and Activities-specific Balance Confidence were used to assess handicap and the level of balance confidence. RESULTS Frequency of falls post treatment was statistically different among groups (P=0.0001); The Berg Balance Scale showed an overall statistically significant difference (P=0.0008) among groups. Change pre-post scores were 6.7, 4.6 and 0.8 points for groups 1, 2 and 3. Dynamic Gait Index showed an overall near statistically significant difference among groups (P=0.14), with change pre-post scores of 3.85, 1.6 and 1.75 points for groups 1, 2 and 3; after the exclusion of drop-outs a statistically significant difference was observed (P=0.04). The self-administered tests (Activities-specific Balance Confidence and Dizziness Handicap Inventory) did not show clinically relevant improvements. CONCLUSIONS Balance rehabilitation appeared to be a useful tool in reducing the fall rate and improving balance skills in subjects with multiple sclerosis. Exercises in different sensory contexts may have an impact in improving dynamic balance.