This paper proposes a group decision framework to model the dependency structure of a complex system based on expert surveys. The proposed framework is designed to systematically populate a dependency matrix whose entry indicates the degree that its row element is influenced by its column element. The degree of dependency is assessed based on two different perspectives—of the influencing element and of the influenced element—and the assessments with the two perspectives are separately compiled and compared in the initial round survey. Entries with large assessment gaps are classified as discussion items and a group decision procedure to reduce the gaps in these items is applied in the consensus round survey. A case study for dependency modeling of critical infrastructures of South Korea was conducted using the proposed framework and its effectiveness was demonstrated. The analysis on the results of the case study discovered an insight that the opinions from the influenced elements are more convincing and respected in the consensus round than those from the influencing elements.