Fluctuating cognition is evidenced in different forms of dementia and is accompanied by electroencephalographic (EEG) abnormalities. The authors hypothesize that cholinesterase inhibitors are effective mostly in patients with fluctuating cognition. Twenty-three patients affected by mild dementia with similar scores on Mini-Mental State Examination (MMSE), Alzheimer Disease Assessment Scale-cognitive subscale (ADAS-cog), and Unified Parkinson's Disease Rating Scale evaluation were classified in a group with fluctuating cognition (n = 11) and a group of nonfluctuators (n = 12). All patients were assigned randomly to the branches of a double-blind crossover study of donepezil (DPZ), a 5 to 10-mg dose, versus vitamin E, a 2000 IU dose, for 30 days. MMSE, ADAS-cog, University of California at Los Angeles Neuropsychiatric Inventory (NPI), quantitative EEG, P3 event-related potentials, choice reaction time variability (CRTV) were assessed at baseline and at the end of treatments. At the end of the crossover study all patients received DPZ for 6 months. The dominant EEG frequency variability, low EEG frequencies amplitude, the P3 latency and jitter, CRTV, and NPI was significantly different in the fluctuating cognition group than the nonfluctuating group at baseline (P < 0.001). Short-term DPZ administration induced a significant increase in MMSE scores, reduction of ADAS-cog and of NPI scores (P < 0.003-0.001), increase of EEG alpha activity and reductions of P3 latency and jitter, dominant frequency variability and CRTV (P < 0.009-0.001) in the fluctuating cognition group, and significant increases of MMSE scores (P = 0.03) and a decrease of P3 jitter and dominant frequency variability (P < 0.034-0.041) in the nonfluctuating group. Short-term DPZ effects differed significantly between fluctuating cognition and nonfluctuating patients (0.001). Significant effects of the 6-month observation were observed only in fluctuating cognition patients. Logistic analysis showed that P3 latency predicts the effect of DPZ (P = 0.04, P < 0.01) in the crossover study, and CRTV predicts the effect at the 6-month follow-up.