Laser detection and ranging (ladar) range images have attracted considerable attention in the field of automatic target recognition. Generally, it is difficult to collect a mass of range images for ladar in real applications. However, with small samples, the Hughes effect may occur when the number of features is larger than the size of the training samples. A random subspace ensemble of support vector machine (RSE-SVM) is applied to solve the problem. Three experiments were performed: (1) the performance comparison among affine moment invariants (AMIs), Zernike moment invariants (ZMIs) and their combined moment invariants (CMIs) based on different size training sets using single SVM; (2) the impact analysis of the different number of features about the RSE-SVM and semi-random subspace ensemble of support vector machine; (3) the performance comparison between the RSE-SVM and the CMIs with SVM ensembles. The experiment’s results demonstrate that the RSE-SVM is able to relieve the Hughes effect and perform better than ZMIs with single SVM and CMIs with SVM ensembles.