A standardized neuropsychological test battery was administered to 167 patients with different forms of mild-to-moderate dementia: probable Alzheimer dementia (AD: n = 49), multi-infarct dementia (n = 43), idiopathic Parkinson disease with dementia (n = 35), depressive pseudodementia (n = 26), and progressive supranuclear palsy (n = 14). Results obtained were used (a) to analyze the profiles of cognitive impairment shown by the different dementia groups; (b) to assess the incidence of some neuropsychological patterns that we hypothesized to be more characteristic of AD, in the various groups; and hence (c) to evaluate the reliability of these patterns as diagnostic markers of AD. Four of the patterns investigated were derived from a verbal learning task (Rey's Auditory Verbal Learning test): (1) absence of the primacy effect; (2) tendency to produce intrusion errors during free recall of a word list; (3) absolute decay of memory trace; and (4) tendency to produce false alarms during delayed recognition of the same word list. Two additional patterns were derived from visual-spatial tasks (copying drawings and Raven's Coloured Progressive Matrices): (5) occurrence of the closing-in phenomenon in copying drawings; and (6) tendency to choose globalistic or odd responses in Raven's matrices. Though all the six patterns were somewhat useful for identifying AD patients, no pattern met the criteria of being both highly sensitive and highly specific, which should characterize an ideal marker. In fact, intrusions and false alarms were observed in many AD patients, but also in patients affected by other forms of dementia. The absence of the primacy effect, the closing-in phenomenon, and the absolute decay of memory trace were more specific, but could be observed in only one-third of AD patients. We also computed the number of positive patterns shown by each patient and assumed the presence of two or more patterns as a global index suggestive of a dementia of the Alzheimer type. With this cumulative method, a higher level of sensitivity and specificity was achieved in the identification of AD patients.