Detection of changes in dynamic balance could help identify older adults at fall risk. Walking on a narrow beam with its width, cognitive load, and arm position manipulated could be an alternative to current tests. Therefore, we examined additive and interactive effects of beam width, cognitive task (CT), and arm position on dynamic balance during beam walking in older adults. Twenty older adults (69 ± 4y) walked on 6, 8, and 10-cm wide beams (2-cm high, 4-m-long), with and without CT, with three arm positions (free, crossed, akimbo). We determined cognitive errors, distance walked, step speed, root mean square (RMS) of center of mass (COM) displacement and trunk acceleration in the frontal plane. Beam width decrease progressively reduced distance walked and increased trunk acceleration RMS. Step speed decreased on the narrowest beam and with CT. Arm crossing decreased distance walked and step speed. COM displacement RMS and cognitive errors were not affected by any manipulation. In conclusion, distance walked indicated that beam width and arm position, but less so CT, affected dynamic balance, implying that beam walking has the potential to become a test of fall risk. Stability measurements suggested effective trunk adjustments to control COM position and keep dynamic balance during the task.