Older people tend to walk more slowly than younger people (Dingwell & Marin 2006), but it is not clear if this is different for elderly fallers – nor whether there is a change in the quality of their walk. Our study compared 34 young subjects (mean 28 sd 10 years) 43 elderly subjects (72 sd 5) and 20 elderly fallers (75 sd 4, who self-reported an average of 1.8 trips in the previous year). Walking in a laboratory setting was measured by both CODA mpx30 (Charnwood Dynamics, UK) 3D motion analysis and with bi-axial accelerometers (ADXL202, Analogue Devices, UK). Longer walks (22m) could be measured only by the accelerometers with data acquisition via an iPAQ Pocket PC. Subjects were asked to walk either at their normal speed or to walk slowly. The young walked faster than both older groups and the older non-fallers walked faster than the older fallers (Bonferroni p<0.02). Somewhat surprisingly the older fallers walked much slower during the slow walk trials (34% of normal c.f. 61% for non-fallers), despite same instructions to all participants. Activity, estimated as the SD of acceleration, was measured in the mediolateral and vertical directions at the spine (L5) and ankle. This was directly proportional to walking speed (R² = 0.91) for mediolateral motions but show a quadratic term at higher speeds for vertical movements. An index was thus used of (sd acceleration) / (walking speed) which essentially measures the untidiness of walking. No significant differences were found in this index for movements in the vertical direction. However there was a significant increase only for elderly fallers in mediolateral movements at both spine and ankle positions but only for the self-selected slow walk ( Spine L5: young 0.11 se 0.01, elderly 0.11 se 0.01, fallers 0.20 se 0.02 g / m/s). The mediolateral movements at the ankle were further analysed by chopping the record into mid-swing and terminal-swing phases, each 15% of the stride time. This showed that the larger part of the increase in variability of movements in the fallers occurred during the terminal-swing phase. Although the records from accelerometers are a compound of changes in linear acceleration and angular changes in gravitational acceleration, our comparison with CODA measurements show they give an effective way of quantifying variability of movements. After correcting for changes in walking speed both elderly and young subjects had the same variability. However fallers were characterised by having a considerably larger variability index for mediolateral movements at slow walking speeds. This was due in part to them choosing to walk at a much lower speed and part from the increased sd of acceleration. The latter appears to be substantially associated with the terminal-swing phase, i.e. uncertainty of foot placement.