The many tasks used for clinical memory assessment have not been compared systematically for their usefulness in differentiating dementia and depression in old age. The aim of this meta-analysis was to identify those attributes of memory tasks that show high discriminative power. Eighty-nine effect-sizes were calculated out of 16 publications directly comparing demented and depressed patients. Outliers in the effect-size distribution (5% of the highest values) were excluded. The groups could be significantly better differentiated by delayed retrieval tasks rather than immediate retrieval tasks. Tasks with distraction before retrieval reached higher effect sizes than retrieval tasks without distraction. Tasks of high-capacity demand differentiated the groups significantly better than tasks of moderate and low demand. Effect-size magnitude was not influenced by patient characteristics except severity of dementia. Thus, demented and depressed patients may best be differentiated by a memory task that uses delayed retrieval with distraction.