In order to improve the stability of the feature and the false reject rate of discriminator for SAR ATR,we propose a new target discrimination method based on region covariance matrix feature and an iterative manifold support vector data d escription (SVDD).A new discrimination feature based on the region covariance matrix is co nstructed, which fuses several single texture features and considers the correlations among these single features.These features can be separated easily in our experiments and further feature selection method is no longer required.A new iterative manifold SVDD discrimina tor is designed by combining the SVDD classification principles with the manifold struc ture of the covariance matrix feature space,which choose the center of the hypersphere in SVDD instead of the Karcher mean as the base point by a novel iterative algorithm.Ex perimental results on RADARST-2data demonstrate the effectiveness of the proposed method.