Signal analysis and pattern recognition methods for brain functional near-infrared spectroscopy (fNIRS) are especially important for its research and applications in the field of cognitive science. The traditional statistical feature extraction method for fNIRS is briefly reviewed and a new feature extraction method based on the principle of multivariate graph representation is proposed. The pattern recognition experiments based on both methods are conducted and compared. The experimental results indicate that the feature extraction method for fNIRS signals based on the multivariate graph representation principle can be used for signal analysis and visualization, which offers a new approach for the analysis of fNIRS signals.