The identification of the distinct conformation classes of a molecule is a common and often crucial step in establishing structure-function relationships. Many different methods have been suggested for that purpose which differ in their choice of a (dis)similarity measure and clustering algorithm. The present study discusses and analyzes these issues, proposing a method based on principal component analysis (PCA), which is applied to conformations obtained from molecular dynamics (MD) simulations of an arginylglutamate repeat. Simulations are done at different pH values, using both standard MD and constant-pH MD methods, with the peptide displaying a very high conformational variety. The conformational analysis starts with a comprehensive comparison of different sets of conformational coordinates and of their ability to preserve structural similarity between conformations. The selected set of conformational coordinates is then used to investigate the preservation of structural similarity after PCA transformation, concluding the need of using a multidimensional conformation space. This conformation space is then used to derive a multidimensional probability density and the corresponding energy landscape. The application of a simple cutoff algorithm to the resulting multidimensional landscape is then shown to produce a consistent set of distinct and homogeneous conformation classes. Overall, this methodology provides an efficient way to identify the major conformation classes of a molecule in a way that directly reflects the density of states in the multidimensional conformation space, contrasting with the more heuristic nature of standard clustering methods.