The great advances in the field of neuron tracing have made possible a high availability of free-access data in the Internet, which encourages the realization of automatic classifications. The increase of neuronal reconstruction databases makes the manual classification of neurons a time-consuming and tedious task for human experts. Classification by human experts is also prone to inter- and intra-analyst variability due to the process’ inherent subjectivity. In this context, the need arises to find new descriptors having discriminative properties which allow separating the various neuron classes, and this constitutes currently an open problem. Such descriptors would contribute to improve the results of automatic classification. In this study the attention is focused on the use of new morphological features in supervised classification of traced neurons. Furthermore, we present a comparative analysis of different supervised learning algorithms oriented to the classification of reconstructed neurons. The results were validated using a non-parametric statistical test and show the usefulness of the proposed solution.