Background: Repeated administration of psychometric instruments frequently results in a higher score on retesting, the so-called training effect. Yet, a training effect has been poorly considered in longitudinal studies of cognitive changes in older persons. Methods: We investigated the presence, magnitude and potential adjustments for training effect in the older participants of the Systolic Hypertension in the Elderly Program (SHEP). SHEP evaluated the cognitive status effects of a diuretic-based treatment of isolated systolic hypertension versus placebo. Changes in the short Comprehensive Assessment and Referral Evaluation (short-CARE) questionnaire score, from baseline through 4 years of follow-up, were assessed in 4,718 participants. In this study, we used two regression techniques to adjust data for the training effect. Results: In both study groups, a training effect was evident as a progressive improvement in the short-CARE score throughout year 1. Thereafter, cognitive scores tended to deteriorate, more in the placebo than in the active treatment group (p = 0.055). When follow-up scores were adjusted based upon baseline data, the difference between the study groups reached statistical significance (p = 0.019), but the apparent overall trend towards deterioration in cognitive score was no longer observed. Adjustment of baseline data preserved this apparent temporal course, but did not improve the discrimination between the two study groups (p = 0.165). Conclusions: In SHEP, repeated cognitive assessments were likely biased by a training effect which could be only partially corrected by statistical techniques. Studies of changes in the cognitive status of older persons should be designed appropriately to estimate and minimize the consequences of a training effect in follow-up data.