Dual-task walking performance is an early marker of dementia. However, there is uncertainty about which measure of the dual-task test is a better marker. The objective of this study was to determine which dual-task measure best differentiates between normal cognition, mild cognitive impairment (MCI) and dementia. Participants (n = 116) were aged ≥60 years attending a cognitive clinic in Melbourne, Australia. Single- and dual-task gait speed were obtained using a 16 metre distance and stopwatch. The cognitive task involved reciting alternate letters of the alphabet sitting and walking. Dual-task interference in gait and cognition was calculated as: single-task-dual-task/single task × 100 and summed to obtain total interference. Multiple linear regression was used to determine differences in single and dual-task measures between those with no cognitive impairment (n = 11), MCI (n = 54) and dementia (n = 51). The mean age of the sample was 76.9 (SD 6.4) years and 48.3% (n = 56) were female. Compared to those with dementia: (a) those with MCI had a higher dual-task letter rate and lower cognitive and total interference (all indicate better performance) (p < .05) and (b) those with no cognitive impairment had a higher single- and dual-task letter rate (both indicate better performance) (p < .05). There were no differences between those with no cognitive impairment and those with MCI (all p > .05). In a cognitive clinic, measurement during dual-task walking differentiated those with dementia from those with MCI or no cognitive impairment. However, differences appear to be driven by performance on the cognitive, rather than the gait task.