Since impaired mediolateral balance can increase fall risk, especially in the elderly, its quantification and training might be a powerful preventive tool. We propose a visual tracking task (VTT) with increasing frequencies (.3-2.0Hz) and with center of pressure as visual feedback as an assessment method. This mediolateral balance assessment (MELBA) consists of two tasks, tracking a predictable target signal to determine physical capacity and tracking an unpredictable target signal to determine sensorimotor integration limitations. Within and between sessions learning effects and reliability in balance performance descriptors in both tasks were studied in 20 young adults. Balance performance was expressed as the phase-shift (PS) and gain (G) between the target and CoP in the frequency domain and cut-off frequencies at which the performance dropped. Results showed significant differences between the MELBA tasks in PS and G indicating a lower delay and higher accuracy in tracking the predictable target. Significant within and between sessions learning effects for the same measures were found only for the unpredictable task. Reliability of the cut-off frequencies at which PS and G performance declined and the average values within cut-off frequencies was fair to good (ICC .46-.66) for the unpredictable task and fair to excellent for the predictable task (ICC .68-.87). In conclusion, MELBA can reliably quantify balance performance using a predictable VTT. Additionally, the unpredictable tasks can give insight into the visuomotor integration mechanisms controlling balance and highlights MELBA's potential as a training tool.