BACKGROUND: Distinguishing individuals with dementia from those with mild cognitive impairment (MCI) might demand a neuropsychological evaluation. METHODS: One hundred and thirty-one consecutive referred elderly patients (37 clinical-controls, 41 with amnestic MCI and 53 with possible/probable AD) were diagnosed with a comprehensive (full) neuropsychological battery, MRI and clinical data. All of the results were blindly coded and evaluated latter with a subset of the tests to reclassify the subjects as MCI, dementia or clinical-control. Agreement rates between both batteries were calculated. We also used ROC curves to establish the sensitivity and specificity of the brief battery for discriminating (i) clinical-control individuals from a group dementia and MCI patients; (ii) individuals with dementia from individuals without dementia; (iii) clinical-control individuals from a group of MCI. We compared performance of the three groups on all full battery tasks. RESULTS: All neuropsychological tests showed differences between clinical-control and dementia groups. The comparison between MCI and the other groups mainly showed memory differences. Agreement between brief and full batteries was substantial (kappa = 0.805). Analyses with ROC curves showed good sensitivity and specificity to discriminate non-demented (clinical control plus MCI groups) and AD group and also to discriminate clinical-control individuals from individuals with cognitive decline (MCI plus AD group). However, sensitivity and specificity significantly decreased when brief battery was tested to discriminate only normal and MCI diagnosis. DISCUSSION: The use of a brief battery might not be indicated to discriminate MCI and clinical-control individuals, but its use might be adequate to discriminate less specific groups (demented versus non-demented and pathological [dementia and MCI] and non-pathological [clinical-control] groups).