In this paper a modern statistical method is applied to an old cell classification and identification problem in the central nucleus of the inferior colliculus. In a recent computer-based reconstruction study of Golgi-impregnated neurons in the rat, two types of cell with flattened dendritic arbors, flat (F) and less flat (LF), were defined. Both types contributed to the anisotropic and laminar pattern of the nucleus. The classification was based on five morphological features of complete dendritic arbors, two assessed visually and three numerically. With respect to the latter criteria, the two types were classified by preselected cut-off values. The distinction of the two types was supported, among other things, by a prevailing spatial segregation into laminar and interlaminar compartments. The cell sample was too small, however, to validate the classification and segregation definitively. In the present study, the classification is tested by the partial least squares regression method which is independent of the preselected cut-off values, and is able to handle small sample sizes and interdependent variables. In the plots, the F and LF cells are clearly separated into two distinct clusters, strongly supporting the distinction of the two types. The different density of the two clusters shows that the F cells are more homogeneous that the LF cells. The relative importance of the classification criteria is also evaluated. The three-dimensional (3D) inspection and the 3D convex hull-based form factor were found to be the most powerful criteria for identifying the two cell types, while the 2D evaluation of camera lucida drawings, a standard method in neuroanatomy, proved to have the least predictive value.