Background: Levels of physical activity change throughout the year. However, little is known to what extent activity levels can vary, based on accelerometer determined sedentary and physically-active time. The aim of this longitudinal study was to examine older adults' activity changes from a non-snowfall season to a subsequent snowfall season, with consideration of the co-dependence of domains of time use. Methods: Participants were 355 older Japanese adults (53.1% women, aged 65-84 years) living in a rural area of heavy snowfall who had valid accelerometer (Active style Pro HJA-750C) data during non-snowfall and snowfall seasons. Activity was classified as sedentary behavior (SB), light-intensity PA (LPA), and moderate-to-vigorous PA (MVPA). Compositional changes from the non-snowfall to the snowfall season were analyzed using Aitchison's perturbation method. The ratios of each component in the composition, such as [SBsnow/SBnon-snow, LPAsnow/ LPAnon-snow, MVPAsnow/MVPAnon-snow] for seasonal changes, were calculated and were then divided by the sum of these ratios. Results: In men, the percentages of time spent in each activity during the non-snowfall/snowfall seasons were 53.9/64.6 for SB; 40.8/31.6 for LPA; and 5.3/3.8 for MVPA; these corresponded to mean seasonal compositional changes (ΔSB, ΔLPA, ΔMVPA) of 0.445, 0.287, and 0.268 respectively. In women, the percentages of time spent in each activity during the non-snowfall/snowfall seasons were 47.9/55.5 for SB; 47.9/41.0 for LPA; and 4.2/3.5 for MVPA; these corresponded to mean seasonal compositional changes (ΔSB, ΔLPA, ΔMVPA) of 0.409, 0.302, and 0.289 respectively. The degree of seasonal change was greatest in men. Conclusions: In older adults, activity behaviors were changed unfavorably during snowfall season, particularly so for men. The degree of seasonal change was greatest for SB. Development of strategies to keep rural older adults active during the snowfall season may be needed for maintaining a consistently-active lifestyle for their health.