With the widespread use of software systems in the modern society, reliability of these systems have become as important as the functionality they provide. Building reliability into the software development process thus becomes critical for cost effective development and quality assurance. Existing reliability models (applied in post-implementation phases) may not be suitable to address reliability analysis at the software architecture level, as they often rely on implementation-level artifacts. In this paper, we present a framework for classifying reliability models based on their applicability to architectural artifacts, and assess several representative approaches based on the proposed classification. This study highlights several areas for future research.