We examined the influence of using different epoch lengths on the classification accuracy of laboratory-controlled sedentary behaviour (SB), and free-living total time and time spent in bouts of SB and moderate-to-vigorous physical activity (MVPA), in children and adolescents. We used two studies including accelerometer-derived data of: 1) controlled activities, i.e. seven sedentary, one standing and one dancing (n = 90); 2) free-living activities (n = 902). For the controlled-activity data, we calculated percentages of time classified as SB and MVPA. For the free-living data, we calculated medians (25th-75th percentiles) of total time and time spent in bouts of SB and MVPA. Applying 8counts/5seconds, 25counts/15seconds and 100counts/60seconds for SB on controlled-activity data revealed respectively (1) 92-96%, 89-99% and 98-100% of sedentary time accurately classified as SB (activity- and age-dependent); (2) 91-98%, 88-99% and 97-100% of standing time classified as SB (age-dependent); (3) 25-37%, 20-25% and 25-38% of dancing time classified as SB (age-dependent). Using longer epochs, children's total time in SB and MVPA decreased while time accumulated in bouts of SB and MVPA accumulated in bouts increased. We conclude that a 60-second epoch seems preferable when the aim is to classify sedentary behaviour, while a shorter epoch length is needed to capture children's short bursts of MPVA. Furthermore, we should be aware that a longer epoch results in averaging of intensities to the middle category.