In recent years, we have witnessed an increasing interest in producing new quantum chemical methods oriented toward a more and more refined view of the molecular world. Coupled to the explosive development of computers and computing technology, this progress has resulted in the production of an amazing wealth of new data. In general, quantum chemical methods are tested over various collections of atomic/ molecular properties. The evaluation of their predictive capability is still a major issue in computational quantum chemistry. Ransils [1] early paper on the subject bears the eloquent title How good is good agreement with experiment: Evaluating the reliability of quantum-mechanically calculated observables. The possibility of controlling the behavior of methods has been examined early enough [2,3].Sometime ago we proposed an information theoretic approach to the classification of quantum chemical methods. The method was initially applied to the evaluation of the quality of the performance of basis sets on an arbitrary collection of molecular properties, with the FH molecule as a test case [4]. It was further tested on OH [5], BH, CH, NH [6], and the triatomic H 2 O [7] molecule. It was also used by other researchers on a variety of problems as the performance of various types of basis sets in studies on helium [8,9] and copper [10]. Later, the method was extended to treat simultaneously a group of molecules and was tested on a collection of properties for the sequence CH 4 , NH 3 , H 2 O, FH, and Ne [11]. A program for the automatic application of this method has been written and made available to the community of computational chemists by Sordo [12].A most essential element in the evaluation of the performance of quantum chemical methods is the quantification of their relative merit. This quantification always refers to a well-defined collection of atomic/molecular properties. In order to achieve this specific goal, we introduced a new methodology that relies on generalized metrics, graph theory, and pattern recognition techniques [13]. This methodology has been subsequently applied by us [14][15][16] and other authors [17-19] to various problems. Statistical Modelling of Molecular Descriptors in QSAR/QSPR. First Edition. Edited