We analyzed short-term variation of Mini-Mental State Examination (MMSE) and Information-Memory-Concentration (IMC) Test scores in 39 patients with Alzheimer's disease (AD), tested four times over 6 weeks. Although analysis of variance had failed to show a significant "learning" effect or other trends, we reexamined the data using repeated measures models, with and without a learning effect. In the model without a learning effect, mean MMSE scores decreased minimally and mean IMC scores decreased by 0.84 points over 6 weeks. In the model that allowed a potential learning effect between the first and second test sessions, scores increased significantly, by 1.12 +/- 0.47 points for the MMSE and 1.04 +/- 0.43 points for the IMC Test. Patients' test scores predicted from the models had less variability than did their raw scores. The short-term practice effect, although small, should be considered in interpreting changes in scores, especially in therapeutic studies in AD.