Accelerometers allow sensitive and inexpensive measurement of dynamic postural movements under natural conditions of walking (Moe-Nilssen 1998b; Kavanagh 2004). However, the best approach to analysis of these data is not clear. This preliminary study investigated variability in unconstrained walking in young and older people using a protocol approved by King’s College Research Ethics Committee. Ten subjects, five healthy older adults (aged 71±4years, BMI 24.2±1.8, ±SD) and five younger adults (aged 27±6years, BMI 24.6±1.8), were compared. Vertical and mediolateral accelerations at the ankle during an 18 metre walk were recorded using ADXL250 accelerometers and DataqCF2 pocket PC data collection at 300Hz. Stride analysis using feature filtering allowed the total variance of the acceleration to be separated into variance of the mean stride and the residual variance, i.e. the deviations of each stride from the mean stride. A comparison was also made of the effect of transforming the timescale of each stride to a fixed number (200) of points: this reduced the residual SD of mediolateral accelerations by 22% and vertical accelerations (which have sharper transients) by 38%. Vertical acceleration SD of the ankle (Table 1) was somewhat lower in the elderly (by 12%) but the residual SD was much less (35%). Side swing, i.e. the mediolateral SD of acceleration, was also lower in elderly (by 24%) but its residual was not significantly further reduced. This shows, as expected, that vertical power in the elderly is reduced during walking but also that the strides become more stereotypical. In contrast the stride-by-stride variations in lateral movements (the residuals) remain high in the elderly. The high variability of lateral movements probably reflects the neural guidance of the legs during walking and may thus provide a useful index of control of balance. Further refinement of the index needs to relate this to the phase of the stride and its temporal relation to changes in body roll.