Variability and stability of walking of eight transfemoral amputees and eight healthy controls was studied under four conditions: walking inside on a smooth terrain, walking while performing a dual-task and walking outside on (ir)regular surfaces. Trunk accelerations were recorded with a tri-axial accelerometer. Walking speed, mean and coefficient of variation of stride times (ST) and the root mean squares (RMS) of trunk accelerations was calculated. Gait variability and stability were quantified using measures derived from the theory of stochastic dynamics. Regularity was indexed using the sample entropy (SEn) and the scaling exponent α derived form Detrended Fluctuations Analysis. Local stability (LSE) quantified gait stability.Walking speed was lower, but ST variability was not different for amputees than controls. RMS of medio-lateral accelerations was higher for amputees; SEn was higher, implying less predictable accelerations, and LSE higher, indicating decreased stability. The largest condition effect was present for walking outside: trunk RMS increased and LSE decreased.Differences in walking between amputees and healthy controls and their responses to perturbations revealed themselves in the magnitude, variability and stability measures of trunk accelerations. These results imply that quantifying the dynamical structure of trunk accelerations can differentiate between groups with different walking abilities and between conditions of increasing difficulty and may therefore provide a useful diagnostic tool. © 2010 IPEM.