Standardized objective measures of human performance have been introduced in clinical and epidemiologic studies of older populations. Reliability of these measures has usually been estimated by comparing two measures obtained in the same person. However, no information is available on variability of multiple measures collected serially over short time intervals. This study uses data from the Weekly Disability Study, a component of the Women's Health and Aging Study, to describe fluctuations in physical performance over multiple, consecutive time intervals. Walking speed was measured weekly over a 6-month period in 99 older women affected by mild to severe disability. Overall, 2120 observations were explored using techniques developed for the analysis of repeated measures. Results showed that the correlations between observations in the same person were inversely related to their separation in time. The decay in the autocorrelation function was steeper in the least disabled. However, even with 20-week separations in assessments, correlations remained above 0.6 in all age and severity of disability subgroups. Changes over time in performance differed somewhat between disability subgroups, but the relative performance across subgroups remained stable over the entire course of the study. A clear learning effect was found only in those in the middle disability subgroup. Results support the utilization of repeated measures of physical performance in research that evaluates older persons over time.