Multilinear analysis provides a powerful mathematical framework for analyzing synthetic aperture radar (SAR) images resulting from the interaction of multiple factors like sky luminosity and viewing angles, while preserving their original shape. In this paper, we propose a multilinear principal component analysis (MPCA) algorithm for target recognition in SAR images. First, we form a high order tensor with the training image set and we apply the higher-order singular value decomposition (HOSVD) to reveal patterns and dependencies between images. The HOSVD of this training image tensor is also used for compressing the data and removing background noise. Then, a multilinear projection algorithm exploiting the calculated HOSVD is used to classify an unknown target in a SAR image. This multilinear projection that leads to a nonlinear optimization problem is carried out in an iterative way by applying the alternate least squares (ALS) algorithm which solves a linear projection subproblem at each iteration. The estimated feature vector associated with the mode-class is then used for recognition. Tests with a true SAR image database illustrate very good classification performance of the proposed MPCA-based method while providing a very high compression rate.