Abstract: A novel method for Human face recognition based on two dimensional nonparametric discriminant analysis is proposed. Traditional LDA-based methods suffer from some disadvan-tages such as small sample size problem(SSS), course of a dimensionality, as well as a fundamental limitation resulting from the parametric nature of scatter matrices, based on the Gaussian distribution assumption. To address the problem, a new two dimensional nonparametric discriminant analysis is proposed and a new formulation of scatter matrices is given. Experimental results indicate the robustness and accuracy of the proposed method.