Background and Purpose: The backward walk test (BWT) has been used to evaluate the balance, gait, and fall risk for older adults, but its psychometric properties in older adults with dementia have not been investigated. This study aims to examine the test-retest and interrater reliability, construct and known-group validity, and absolute and relative minimal detectable changes at the 95% level of confidence (MDC95) of the BWT in older adults with dementia. Methods: This study was a cross-sectional study with repeated measures. Thirty older adults with a mean age of 83.3 years and a diagnosis of dementia who were able to walk backward independently for at least 3 m were recruited from day care and residential care units. The BWT was conducted on 3 separate testing occasions within 2 weeks under 2 independent raters using a modified progressive cueing system. The 10-m walk test (10MWT), Berg Balance Scale (BBS), and Timed Up and Go test (TUG) were used to assess the gait, balance, and mobility performances of the participants. Results and Discussion: The BWT had excellent test-retest reliability—intraclass correlation coefficient (ICC) = 0.96—and interrater reliability (ICC = 0.97-0.97) in the participants. Moderate correlations between the BWT and BBS (Spearman's ρ = 0.60) and strong correlations between the BWT and 10MWT (ρ= 0.84) and TUG (ρ=−0.82) were found. The BWT could distinguish between the participants who ambulated with walking aids and those who did not (P = .005). The participants who had experienced a fall in the past year did not differ significantly in the BWT compared with those who had not fallen (P = .36). The absolute and relative MDC95 of the BWT in the participants were 0.10 m/s and 39.3%, respectively. Conclusions: The BWT is reliable and valid in assessing balance and gait performances in older adults with dementia. Further investigation is needed to determine whether the BWT can identify those with an increased risk of falls.