The Alzheimer's disease (AD) has a wide spectrum of symptoms, ranging from cognition dysfunction to behavior disturbances and functional impairment. The evoked cerebral potentials by specific paradigms are useful for disclosing neuropsychological activities. The evolution of AD is accompanied by progressive cognitive impairment which may result in a difficulty to recognize or comprehend gestures. In the present study, a visual tool-using gesture paradigm was employed to assess the cognitive functions of 16 probable AD patients, 17 subjects mild cognitive impairment (MCI), and 17 age-matched control subjects. Each subject was conducted by visual stimuli by a series of pictures, each displaying randomly a gesture with correctly or incorrectly using a tool. The P300 amplitude was further used as a parameter to build classifiers based on support vector machine.