Classical principal component analysis (PCA)aims to find low dimensional linear subspace spanned by principal eigenvectors through maximizing variance derivation. The objective of robust L1-norm PCA (L1-PCA) is to maximize the L1-norm variance in feature space. One downside of L1-PCA is that finding a global solution of L1-norm PCA is still intractable. In this talk, we transform L1-PCA into an equivalent integer quadratic program, which is NP-complete.We present a simple semidefinite programming (SDP)relaxation for obtaining a nontrivial upper bound on the optimal value of the equivalent problem, from which an approximate solution is obtained.The effectiveness of the method is shown for numerical probleminstances of various sizes.