In this paper, the problem of classifying nonhomogeneous man-made targets is investigated by performing a macroscopic and detailed target analysis. The Cloude-Pottier H/ αML decomposition is used as a starting point in order to find orientation-invariant feature vectors that are able to represent the average polarimetric structure of complex targets. A novel supervised classification scheme based on nearest neighbor decision rule is then designed, which makes use of the feature space. A validation process is performed by analyzing experimental data of simple targets collected in an anechoic chamber and airborne EMISAR images of eight ships. Three classification robustness performance indicators have been evaluated for each feature vector by performing the leaves-one-out-method described by Mitchell and Westerkamp. The robustness of the classifier has been tested with respect to the ability to reject unknown targets and to correctly identify known targets.