In the last decades, a number of developments have made global optimization of large multi-dimensional design option spaces possible. Such developments include the increase in computing power, emergence of sophisticated optimization algorithms, and new techniques for the derivation of computationally highly efficient meta-models. Along with their promise, such developments also involve a number of potential drawbacks. For one thing, meta-models occasionally fail to capture the behaviour of "non-conventional" and complex designs. Another critical problem pertains to the potentially opaque nature of large-scale global optimization exercises, which make them less amenable to provision of intuitively graspable support in a naturally iterative design process. In this context, this research explores the potential of a novel approach toward iterative global optimization of locally optimized attribute clusters of building design solutions. Thereby, clusters of design space attributes (i.e., sets of ontologically cognate aspects of designs) that are comprehensible to typical building designers as a compound yet coherent aspects of a design are made subject to multiple passes of local simulation-assisted optimization runs. Hence, instead of allocating an individual dimension to each and every variable of a complex design within the context of a single-pass global optimization campaign, multiple iterative optimization steps target coherent clusters of such attributes and pursue those until the overall design meets the expected performance (or until further performance improvement is not forthcoming). The dissertation reports on several tests of this approach, documenting the methods advantages (i.e., use of original simulation models instead of meta-models as well as iterative, transparent, and intuitive navigation of the design space). The performance of the implementations of the proposed approach via optimization case studies, which contained different system operation options are illustrated (e.g., random cycling between attribute clusters versus predefined sequences as well as different complexities of the buildings). The proposed method delivers optimized solutions that are as far as the values of the energy performance indicators and the associate cost function are concerned virtually indistinguishable from those of a reference one-shot global optimization run. However, in this approach the results are not only obtained faster, more efficiently and more accurately, but also via a transparent, traceable, and designer-friendly process.