The objective of this work is to assess the possibility of using (Electroencephalogram) EEG for communication between different subjects. Here EEG signals are recorded from 13 subjects by inducing the subjects to imagine the English vowels ‘a’, ‘e’, ‘i’, ‘o’ and ‘u’ through visual stimulus. These recorded signals are then processed to remove artifacts and noise. Common features: Average power, Mean, Variance and Standard deviation are computed and classified using bipolar neural network. This method yields maximum classification accuracy of 44%. The result shows that EEG has some distinctive information for across subject classification.