We applied latent class analysis (LCA) to a set of neuropsychological data with the aim of corroborating the three cognitive profiles of mild cognitive impairment (MCI) described in the literature, namely: healthy, amnestic, non-amnestic, and multidomain. The ultimate purpose of the LCA was to try to find the underlying classification of MCI and related pathologies by means of the participants’ response patterns, rather than on more classical psychometric criteria, such as the standard deviation of the mean. We computed 547 neuropsychological assessments derived from 223 participants who were assessed annually for three consecutive years. The battery included tests of memory, language, executive function, and praxis. The results obtained by means of LCA, with a four-group solution and using the 40th percentile as the criterion, confirm prior classifications obtained with more questionable psychometric criteria, while providing longitudinal data on the course of MCI and the stability of group assignment over time.