BACKGROUND AND PURPOSE: Angular movements of the pelvis and lumbar spine are thought to play an important role in walking. However, little is known about the amount of unpredictable variability in measurement of these movements during human walking. The aim of the present study was to determine the retest reliability of measuring the angular movements of the pelvis and lumbar spine during unimpaired familiarized treadmill walking. METHOD: Retest reliability for 26 subjects without pathology was determined over a one-week interval. Subjects walked on a treadmill at self-selected or a slower speed while measurements of the three-dimensional angular movements were taken with a computer-based video analysis system. RESULTS: The frontal plane movements of pelvic list and lumbar lateral flexion (relative to the pelvis) could be measured with high retest reliability at both self-selected and slow walking speeds (intraclass coefficient (ICC) (2, 1) > or = 0.81). In contrast, transverse and sagittal plane movements demonstrated moderate reliability at both speeds (0.37 < or = ICC (2, 1) < or = 0.76). Averaging the measurement over six strides resulted in increased observed reliability (self-selected walking speed summary Pearson's r = 0.71, slow walking speed summary Pearson's r = 0.79) compared to taking the measurement based on a single stride (self-selected walking speed summary Pearson's r = 0.63, slow walking speed summary Pearson's r = 0.67). Unlike pelvic and lumbar movements (relative to the pelvis), the measurement of lumbar movements (relative to the global reference frame) appeared to depend on whether subjects were walking at self-selected or slow speeds. CONCLUSIONS: Measurement of pelvic list and lumbar lateral flexion (relative to the pelvis) could be applied with confidence to hypothesis testing about individuals or groups. Movements in the transverse and sagittal planes are unlikely to be appropriate in hypothesis testing about individuals and hence clinical practice, but may still have experimental applications in hypothesis testing about groups.