This cross-sectional study examined the association between objectively measured sleep patterns and body composition in very elderly community-dwelling women. Participants included 191 community-dwelling adults aged ≥ 80 years (mean age: 83.4 ± 2.6 years; age range: 80–92 years). Sleep and physical activity were monitored via accelerometer (ActiGraph GT3X+) during at least five consecutive 24-h periods. Night-to-night sleep pattern variability across all nights of recording was assessed using standard deviations (SDs). Body composition was assessed using dual-energy X-ray absorptiometry. Simple and multivariable linear regression analyses were performed. The mean number of nights with usable actigraphy data was 7.3 ± 1.3. On average, participants went to bed at 22:57 hours (SD: 1.11 h) and rose from bed at 6:27 hours (SD: 1.01 h). Night-to-night bedtime, sleep duration and sleep timing mid-point variations correlated slightly with the percentage body fat and percentage lean mass (P < 0.05). Multiple linear regression analysis revealed significant associations of night-to-night bedtime variations and inconsistent sleep–wake patterns with all body composition indices after adjusting for potential confounding factors, including mean nightly sleep duration, self-reported nap duration and daily physical activity. After further adjusting for night-to-night wake time, sleep timing mid-point and sleep duration variations, greater bedtime variability remained associated significantly with all body composition indices except lean/fat mass ratio. Inconsistent sleep–wake patterns were associated independently with an increased fat mass and decreased lean mass among very elderly women. These findings suggest that in most elderly individuals, sleep patterns might be an important modifiable factor associated with obesity and sarcopenia development.