Security system architecture governs the composition of components in security systems and interactions between them. It plays a central role in the design of software security systems that ensure secure access to distributed resources in networked environment. To this end, the security system must not only make constituent components work together, but also ensure that the components as a whole behave consistently and guarantee certain end-to-end properties. One such critical property is that the system as a whole must consistently assure security policies that it supposes to enforce. However, there is currently no rigorous and systematic ways to predict and assure such critical properties in security system design. In this paper, a systematic approach is introduced to address the problem. We present a methodology to model security system architectures and to verify whether required security constraints are assured by the composition of its components. We introduce the concept of security constraint patterns, which formally specify the generic form of security policies that all implementations of the system architecture must enforce. The analysis of the architecture is driven by the propagation of the global security constraints onto the components in an incremental process. We show that our methodology is both flexible and scalable. It is argued that such a methodology not only ensures the integrity of critical early design decisions, but also provides a framework to guide correct implementations of the design. We demonstrate the methodology through a case study, in which we model and analyze the architecture of the Resource Access Decision (RAD) Facility, an OMG standard for application-level authorization service.