Recent advances in neuroimaging demonstrate the potential use of functional near infrared spectroscopy (fNIRS) in the field of brain machine interface. An fNIRS uses light in the near infrared range to measure brain surface hemoglobin concentrations to determine a neural activity. The current study presents our empirical results in realizing fNIRS - BCI system. We analyze the hemodynamic responses that are acquired from 4 subjects' frontal cortex using 19-channel fNIRS recordings. A wavelet-neural network methodology is proposed in this study, in order to extract important neural features and to recognize the cognitive tasks. Experimental results demonstrate the potential application of fNIRS for BCI by confirming the best accuracy rate as high as 97% in recognizing the different levels of cognitive tasks. Particularly, we demonstrate efficient way of extracting cognitive neural features by wavelet pre-processing and optimal neural network classifier.