Background: Knowledge is still lacking regarding the preferred method for evaluation of learning in the Rey Auditory Verbal Learning Test (RAVLT). Validity of different methods was examined by the effect size in differentiating diagnostic stages in memory clinic patients versus healthy adults and the strength of association between RAVLT performance and brain atrophy. Method: The study included individuals with dementia (n = 247), Mild Cognitive Impairment (MCI, n = 709), Subjective Cognitive Impairment (SCI, n = 175) and cognitively unimpaired adults serving as healthy controls (HC, n = 102). All patients went through a comprehensive clinical examination and neuropsychological assessment of cognition including episodic memory gauged with RAVLT and brain imaging of medial temporal atrophy, cortical atrophy, and white matter hyperintensity. Results: The standard method for evaluation of learning in RAVLT (summed score over five trials) together with the late learning method (mean of trials 4 and 5) were the two most powerful methods according to group differentiation (discriminant validity). Both methods also showed considerable association with medial temporal atrophy (construct validity). The initial RAVLT performance represented by results on trial 1 and the constant in regression analysis with the power function provided information regarding attention that was important for the separation of SCI and HC. Conclusions: The most favorable clinical utility was indicated by discriminant and construct validity by total learning (standard method) including both attention- and learning-related parts and late learning of RAVLT performance, while theoretical understanding of mental processes involved in RAVLT performance was provided by the distinction between initial versus the subsequent learning performance.