Regression models were developed to explain age-related and total variance in memory and to determine the independent contribution from general processing speed, having taken into account cognitive and noncognitive individual differences. Episodic memory was assessed for 3 tasks in a population-based sample of 951 adults comprising 515 men and 436 women (aged 70-96, M = 77.6, SD = 5.5). Correlations between age and memory accounted for 6%-9% of the variance. Hierarchical multiple regressions showed a reduction in this age-related variance by up to 94%, after entering gender, depression, health, cognitive status, activities, and speed. General processing speed was the major mediator of age-related variance in memory. Although both the age-related variance and the speed-related variance in memory were significantly reduced by prior entry of other individual differences variables for all 3 tasks, speed remained a significant mediator of remembering, and negligible differences in the residual age-related variance were observed by inclusion of other background variables.