Including resilience in an overall systems optimization process is challenging because the space of hazard-mitigating features is complex, involving both inherent and active prevention and recovery measures. Many resilience optimization approaches have thus been put forward to optimize a system’s resilience while systematically managing these complexities. However, there has been little study about when to apply or how to adapt architectures (or their underlying decomposition strategies) to new problems, which may be formulated differently. To resolve this problem, this article first reviews the literature to understand how choice of optimization architecture flows out of problem type and, based on this review, creates a conceptual framework for understanding these architectures in terms of their underlying decomposition strategies. To then better understand the applicability of alternating and bilevel decomposition strategies for resilience optimization, their performance is compared over two demonstration problems. These comparisons show that while both strategies can solve resilience optimization problem effectively, the alternating strategy is prone to adverse coupling relationships between design and resilience models, while the bilevel strategy is prone to increased computational costs from the use of gradient-based methods in the upper level. Thus, when considering how to solve a novel resilience optimization problem, the choice of decomposition strategy should flow out of problem coupling and efficiency characteristics.