The purpose of this study was to compare the efficiency of three different balance training strategies in an effort to understand the mechanisms underlying training-related changes in dual-task balance performance of older adults with balance impairment. Elderly individuals with balance impairment, age 65 and older, were randomly assigned to one of three individualized training programs: single-task (ST) balance training; dual-task training with fixed-priority (FP) instruction; and dual-task training with variable-priority (VP) instruction. Balance control during gait, under practiced and novel conditions, was assessed by calculating the center of mass and ankle joint center inclination angles in the frontal plane. A smaller angle indicated better balance performance. Other outcomes included gait velocity, stride length, verbal reaction time, and rate of response. All measures were collected at baseline and the end of the 4-week training. Results indicated that all training strategies were equally effective (P>.05) at improving balance performance (smaller inclination angle) under single-task contexts. However, the VP training strategy was more effective (P=.04) in improving both balance and cognitive performance under dual-task conditions than either the ST or the FP training strategies. Improved dual-task processing skills did not transfer to a novel dual-task condition. Results support Kramer et al.'s proposal that VP training improves both single-task automatization and the development of task-coordination skills.