Unsupervised classification used the classical hierarchical agglomerative algorithm (HAC) illustrated by an UPGMA tree based on Euclidean distances between shape variables. Supervised classification was performed as a validated one, i.e., each reclassified case did not contribute to the model used to perform the classification. The model was an artificial neural network (ANN) making use of a simple multilayer perceptron (MLP) with a back- propagation algorithm [42]. The method has been recently applied to morphometric data, including outline-based morphometrics [43,44]. Following a process of trial-and-error, a single (hidden) layer composed of three neurons provided the best results. We used as input the total number of variables instead of a subset of their PCs. All specimens were separately classified 10 times, then an average classification score and its standard error were computed. The “accuracy” (Table 2) was the percentage of individuals correctly identified at the end of the procedure and is provided with the standard deviation.