In this prospective longitudinal study, conducted in a large sample of amnestic MCI patients over a three-year period, we investigated the recently advanced proposal that unadjusted test scores obtained at baseline on long-term memory tests are more reliable than age- and education-corrected scores in predicting progression from aMCI to AD. Our experimental sample consisted of 270 aMCI patients who underwent extensive neurological and neuropsychological examinations both at baseline and at the follow-up, conducted at least 3 years later. At the follow-up 80 patients had converted to overt dementia. The predictive capacity of raw, age-corrected, education-corrected and fully corrected scores on RAVLT immediate and delayed recall was compared by examining the area under the ROC curves (AUCs) of all of these scores to assess which (raw or corrected) scores achieves the better reliability in predicting conversion to dementia. The condition (aMCI stable vs converted) was analyzed to assess the odds ratios resulting from a logistic regression on the corrected and uncorrected scores of RAVLT immediate and delayed recall. Even if both in immediate and in delayed recall the ROCs of 'raw scores' were generally higher than the other ROCs on corrected scores, these differences did not reach the level of statistical significance, failing to support the claim that unadjusted test scores are superior to age- and education-corrected scores in predicting progression from aMCI to AD.