OBJECTIVE: Allostatic load, a composite measure of accumulated physical wear and tear, has been proposed as an early sign of physiological dysregulation predictive of health problems, functional limitation, and disability. However, much previous research has been cross sectional and few studies consider repeated measures. We investigate the directionality of associations between allostatic load, self-rated health, and a measure of physical function (walking speed). METHODS: The sample included men and women 60 and older who participated in Wave 2 (2004) and Wave 4 (2008) of the English Longitudinal Study of Ageing (n = 6132 in Wave 2). Allostatic load was measured with nine biomarkers using a multisystem summary approach. Self-rated health was measured using a global 5 point summary indicator. Time to walk 8 ft was used as a measure of function. We fitted and tested autoregressive cross-lagged models between the allostatic load measure, self-rated health, and walking speed in Waves 2 and 4. Models were adjusted for age, sex, educational level, and smoking status at Wave 2 and for time-varying indicators of marital status, wealth, physical activity, and social support. RESULTS: Allostatic load predicted slower walking speed (standardized estimate = -0.08, 95% confidence interval [CI] = -0.10 to -0.05). Better self-rated health predicted faster walking speed (standardized estimate = 0.11, 95% CI = 0.08-0.13) as well as lower allostatic load (standardized estimate = -0.15, 95% CI = -0.22 to -0.09), whereas paths from allostatic load and walking speed to self-rated health were weaker (standardized estimates = -0.05 [95% CI = -0.07 to -0.02] and 0.06 [95% CI = 0.04-0.08]). CONCLUSIONS: Allostatic load can be a useful risk indicator of subsequent poor health or function.