Nettelbeck & Rabbitt (1992) found that measures of speed of performance with low knowledge requirements (four-choice reaction time, inspection time and coding-substitution) accounted substantially for age-related changes among 104 persons aged from 54 to 85 years in a number of more complex cognitive measures reflecting general fluid ability. However, the numbers of words recalled from a list after either a single brief viewing of each word, or following a cumulative learning procedure across four trials, provided an exception to this general trend, leading to the conclusion that some aspects of memory may be independent of mental speed. A follow-up of 82 of the same people 18-20 months later was designed to partition performance in a similar cumulative learning procedure into an initial first recall component and a subsequent learning component. This was accomplished by fitting individual cumulative learning data with a hyperbolic power function which met the theoretical requirement of defining separate initial recall and learning parameters. These parameters were found to be independent and it was concluded that learning involved rehearsal, whereas first recall did not. The hyperbolic power function provided a good account for 92 per cent of individual cases. Analyses which combined Nettelbeck & Rabbitt's (1992) data with new measures confirmed the reliability of these authors' results. Furthermore, it was found that first recall, but not learning, was mediated by processing speed. Learning was relatively unaffected by age-related slowing in mental speed, suggesting that Nettelbeck & Rabbitt's results were the consequence of a strong rehearsal component in their memory tasks. Thus, while mental slowing is clearly one important aspect of cognitive decline during old age, it does not constitute a sufficient explanation for changes in all areas of cognitive functioning. Specifically, age-decline in rate of learning with rehearsal appeared to be independent of slowing in speed of information processing.