Measuring recovery of function may mean testing the same individual many times, a procedure that is inevitably open to improvement due to learning on the specific tests rather than recovery per se. This is particularly likely to be an issue with measures of memory performance. We therefore studied the performance of normal and brain-injured people across 20 successive test sessions on measures of orientation, simple reaction time, forward and backward digit span, visual and verbal recognition, word list learning and forgetting, and on three semantic memory measures, namely, letter and category fluency and speed of semantic processing. Differences in overall performances between the two groups occurred for all tests other than orientation, digit span forward, and simple reaction time, although the tests differed in their degree of sensitivity. The tests varied in the presence or absence of practice effects and in the extent to which these differed between the two groups. Data are presented that should allow investigators to select measures that are likely to optimize sensitivity while minimizing possible confounding due to practice effects. (JINS, 2000, 6, 469–479.)