Neuropsychological test batteries are repeatedly administered to evaluate changes over time or the effects of clinical interventions. Relationships between scores on different tests within batteries are also examined to test models for associations between functional deficits. These comparisons may be misleading unless Test/Re-test reliability for individual tests is satisfactory. Interpretations of repeated measurements also depend on the extent to which improvement with practice varies between tasks and between more and less able individuals. Test/Re-test correlations and practice effects for two neuropsychological test batteries (CANTAB, ISPOCD) and from laboratory tasks commonly used in cognitive assessments of older people were obtained from large groups of healthy elderly. Tests in neuropsychological batteries varied markedly in test/re-test reliability which, in some cases, fell below levels considered methodologically acceptable. Putative measures of 'frontal' or 'executive' function, in which performance may be markedly improved by abrupt discovery of an appropriate strategy, were especially likely to show low reliability. Most tests showed significant practice effects, and on some these are substantial enough to compromise comparisons on repeated testing. On a minority of tests practice effects were counter-intuitive, in that less able showed significantly more gains than more able individuals.