Early detection of balance impairment is crucial to identify individuals who may benefit from interventions aimed to prevent falls, which is a major problem in aging societies. Since mediolateral balance deteriorates with aging, we proposed a mediolateral balance assessment (MELBA) tool that uses a CoM-tracking task of predictable sinusoidal and unpredictable multisine targets. This method has shown to be reliable and sensitive to aging effect, however, it is not known whether it can predict performance on common daily-life tasks such as walking. This study aimed to determine whether MELBA is an ecologically valid tool by correlating its outputs with a measure of mediolateral gait stability known to be predictive of falls. Nineteen community-dwelling older adults (72±5 years) tracked predictable and unpredictable target displacements at increasing frequencies with their CoM by shifting their weight sideward. Response delay (phase-shift) and amplitude difference (gain) between the CoM and target in the frequency domain were used to quantify performance. To assess gait stability, the local divergence exponent was calculated using mediolateral accelerations with an inertial sensor when walking on a treadmill (LDETR) and in daily-life (LDEDL) for one week. Pearson product-moment correlation analyses were performed to determine correlations between performance on MELBA tasks and LDE. Results show that phase-shift bandwidth for the predictable target (range above -90°) was significantly correlated with LDETR whereas phase-shift bandwidth for the unpredictable target was significantly correlated with LDEDL. In conclusion MELBA is an ecologically valid tool for mediolateral balance assessment in community-dwelling older adults who exhibit subtle balance impairments.