Individuals with higher cognitive reserve are more able to cope with pathological brain alterations, potentially due to the application of more efficient cognitive strategies. The extent to which an individual's cognitive performance can be increased by advantageous conditions differs substantially between patients with Alzheimer's dementia (AD) and healthy older adults and can be assessed with the Testing-the-Limits (TtL) approach. Thus, TtL has been proposed as a tool for the early diagnosis of AD. Here, we report the diagnostic accuracy of a memory TtL paradigm to discriminate between AD patients and controls. The TtL paradigm was administered to 57 patients with clinically diagnosed AD and 94 controls. It consisted of a pre-test condition, representing baseline cognitive performance, the presentation of an encoding strategy, and two subsequent post-test conditions, representing learning potential. Receiver operating characteristic (ROC) curves were analyzed for each condition in order to receive optimal cutoff points along with their sensitivity and specificity and to compare the diagnostic accuracy of the conditions. Differentiation between AD patients and controls, indicated by the area under the ROC curve, increased significantly for the TtL post-test and total error scores compared to the pre-test score. The combined error score in the two post-tests could differentiate between AD patients and controls with a sensitivity of 0.93 and a specificity of 0.80. The presented approach can be carried out in 25 minutes and thus constitutes a time- and cost-effective way to diagnose AD with high accuracy.