Age-related decline in processing speed has long been considered a key driver of cognitive aging. While the majority of empirical evidence for the processing speed hypothesis has been obtained from analyses of between-person age differences, longitudinal studies provide a direct test of within-person change. Using recent developments in longitudinal mediation analysis, we examine the speed-mediation hypothesis at both the within-and between-person levels in two longitudinal studies, Longitudinal Aging Study Amsterdam (LASA) and Origins of Variance in the Oldest-Old (OCTO-Twin). We found significant within-person indirect effects of change in age, such that increasing age was related to lower speed, which in turn relates to lower performance across repeated measures on other cognitive outcomes. Although between-person indirect effects were also significant in LASA, they were not in OCTO-Twin which is not unexpected given the age homogeneous nature of the OCTO-Twin data. A more in-depth examination through measures of effect size suggests that, for the LASA study, the within-person indirect effects were small and between-person indirect effects were consistently larger. These differing magnitudes of direct and indirect effects across levels demonstrate the importance of separating between- and within-person effects in evaluating theoretical models of age-related change.