The present study evaluated the effect of caffeine (225 mg) on cognitive performance in young, middle-aged, and old subjects in a placebo-controlled parallel groups design (n=60). Groups were matched for level of education and sex. Positive effects of caffeine, as compared to placebo, were found in middle-aged subjects in the first trial of the word learning test. In contrast, caffeine had negative effects on the speed of searching short-term memory in young subjects. Caffeine had no effect on the intercept, which is an indicator for sensorimotor speed, of a memory scanning task. The middle-aged subjects appeared to regularly consume twice as much caffeine as the young and old subjects. These results were similar to earlier findings in a large population study. Although statistical analyses with habitual caffeine consumption as a covariate did not yield different results, a caffeine withdrawal effect was hypothesized to be responsible for the reduced cognitive performance of middle-aged subjects receiving placebo. The habitual use of large amounts of caffeine by middle-aged subjects may be a means to overcome the age-related decrease in cognitive functioning that is caused by changes in information processing. © 1998 John Wiley & Sons, Ltd.