From his own experience, the author discusses the need of Metallurgy for theoretical models and their limits. The development of simple but less accurate models has provided a semi-quantitative approach of many complex metallurgical phenomena that has proven useful to orientate research as well as to afford interpretation-based prediction. Such models can be validated within a narrow range only. These approximate models could be substituted for by elaborated and comprehensive calculations that would provide reliable predictions with the help of ever more efficient computers. It remains a difficult task to take satisfactorily into account every internal parameter and boundary condition of a global system. Furthermore, any new observation requests another calculation that gives a clear picture but no comprehensive understanding of the underlying phenomena. Combining experiments, interpretation and prediction modeling along with few dedicated calculations may eventually allow further progress in such a complex domain as Metallurgy.