The purpose of the present study was to relate 3D acceleration patterns of the lower and upper trunk during running to running gait cycle, assess the validity of stride duration estimated from acceleration patterns, investigate speed-dependent changes in acceleration, and examine the test-retest reliability of these parameters. Thirteen healthy young men performed two running trials each on a treadmill and on land at three speeds (slow, preferred, and fast). The 3D accelerations were measured at the L3 spinous process (lower trunk) and the ensiform process (upper trunk) and synchronised with digital video data. The amplitude and root mean square of acceleration and stride duration were calculated and then analysed by three-way analysis of variance to test effects of running conditions, device location, and running speed. Bland-Altman analysis was used to evaluate the test-retest reliability. Marked changes in acceleration were observed in relation to foot strike during running. Stride durations calculated from the vertical accelerations were nearly equal to those estimated from video data. There were significant speed effects on all parameters, and the low test-retest reliability was confirmed in the anterior-posterior acceleration during treadmill running and the anterior-posterior acceleration at slow speed during treadmill and overground running.