Background/Aims: Global cognitive scales and meta-analyses thereof are used to appraise therapeutic efficacy over a broad range of disease severity. Clinically, however, different aspects of cognition change in different stages of disease. Methods: Calculation of effect sizes for single cognitive functions on treatment as assessed by the Alzheimer’s Disease Assessment Scale (ADAS-cog), the Mini-Mental-Status Examination (MMSE), and the Severe Impairment Battery (SIB). In these scales, subdomains of ‘cognition’, e.g. memory and language, are represented in different proportions. To exemplify the analysis of ‘cognition’, we used original data of previously published clinical studies with memantine. Results: Depending on dementia severity and on the scale used, the effect size for memory varies between –0.44 and +0.34 and for language between –0.40 and +0.26. Conclusion: Beyond interstudy variance, effect sizes for treatment with antidementia drugs are subject to disease stage, instruments used, and interaction thereof. Therefore, clinical interpretation is necessary to appraise therapeutic efficacy in clinical studies and meta-analyses thereof when patients with different severity are included or different instruments are used. Alternatively, severity-adapted endpoints should be used for appraisal and meta-analysis of therapeutic efficacy.