Industrial catalysts often owe their remarkable activity, selectivity and reliability to many years of gradual improvement and optimization. Such research largely relies on physically/chemically motivated systematic variations of important parameters such as catalyst composition and working conditions, but often there is only moderate emphasis placed on the elucidation of the fundamental reasons for a catalyst's success. This “trial-and-error” approach is chosen not because of any strong reluctance to discover a catalyst's intrinsic workings but, rather, because modern catalysts are extremely complicated systems, making fundamental investigations expensive in time and money and with no guarantee of useful results. One reason for this is that, from the theoretical analysis point of view regarding such complex systems, it can appear almost impossible to distinguish the catalytically important and active aspects from the redundant and passive. The assumption that this division of role can be made, however, lies at the root of most intuitive ideas relating to catalytic activity. In this article, we aim to illustrate that by combining computational approaches with this conceptual division of role much benefit can be derived. Such considerations are based around the general concept of localized active sites, as distinguished from the supporting liquid or solid environments of the catalyst, which are assumed to be relatively inert but not without influence. We will show how this concept can be used as the starting point of modern computational modeling techniques, which can be applied at a number of different levels to heterogeneous catalytic systems, pointing the way to a more efficient approach to catalyst optimization and understanding than trial-and-error. Our account will culminate in one of the most computationally extensive descriptions of an active site yet achieved.