Although traditional software metrics have widely been applied to practical software projects, they have insufficient abilities to measure a large-scale system?s complexity at high level so as to provide an overview of the system for developers. So, an adequate metrics set for large-scale software systems that can comprehensively measure the complexity at various levels is still challengeable. First, we summarize universal properties and implicit limitations of recognized object-oriented metric sets in the face of ever-increasing complexities of modern software systems. Large-scale software systems represent an important class of artificial complex networks. Then, from the perspective of software engineering, the main parameters of complex networks are introduced in detail. Furthermore, we integrate these metrics and parameters into a hierarchical complexity metrics set, which can measure the complexity at different levels of a large-scale software system. Eventually, we prove the feasibility of our metrics set through analyzing the data from a software project.