Impairment of cognitive and working memory after stroke was common. Vascular dementia (VaD) was a prevalent type of dementia that was caused by an impaired blood supply to the brain because of a series of small strokes. Electroencephalogram (EEG) gives information about brain status and activity, so it had a lot of potential to be used in diagnosing people with dementia. Since the EEG signal is extremely non-linear and non-stationary data, traditional Fourier analysis such as Fast Fourier Transform (FFT) that broadens sinusoidal signals cannot describe the amplitude contribution of each frequency value in specific time. Meanwhile, Hilbert Huang Transform (HHT) was based on the characteristic local time scale of the signal, it can efficiently obtain instantaneous frequency and instantaneous amplitude for nonstationary and nonlinear data. In this paper, HHT was employed as feature extraction method to extract the energy features of frequency bands from post stroke patients and healthy subjects. The extracted features were fed into extreme learning machine (ELM) for classifying post stroke patient with VaD and healthy subjects. The results of classification accuracy using HHT as feature extractor and FFT as feature extractor were compared. The mean accuracy of classification using HHT was 59.14%, respectively, while mean accuracy of classification using FFT was 94.4%, respectively, in classifying post stroke patient with VaD and healthy subjects.