BACKGROUND: Gait variables derived from trunk accelerometry may predict the risk of falls; however, their associations with falls are not fully understood. The purpose of the study was to determine which gait variables derived from upper and lower trunk accelerometry are associated with the incidence of falls, and to compare the discriminative ability of gait variables and physical performance. METHODS: This study was a 1-year prospective study. Older people (n = 73) walked normally while wearing accelerometers attached to the upper and lower trunk. Participants were classified as fallers (n = 16) or non-fallers (n = 57) based on the incidence of falls over 1 year. The harmonic ratio (HR) of the upper and lower trunk was measured. Physical performance was measured in five chair stands and in the timed up and go test. RESULTS: The HR of the upper and lower trunk were consistently lower in fallers than non-fallers (P < 0.05). Upper trunk HR, was independently associated with the incidence of falls (P < 0.05) after adjusting for confounding factors including physical performances. Consequently, upper trunk HR showed high discrimination for the risk of falls (AUC = 0.81). CONCLUSIONS: HR derived from upper trunk accelerometry may predict the risk of falls, independently of physical performance. The discriminative ability of HR for the risk of falls may have some validity, and further studies are needed to confirm the clinical relevance of trunk HR.