Architectural stability refers to the extent software architecture is flexible to endure evolutionary changes while leaving the architecture intact. Approaches to evaluate software architectures for stability can be retrospective or predictive. Retrospective evaluation looks at successive releases of a software system to analyze how smoothly the evolution has taken place. Predictive evaluation examines a set of likely changes and shows the architecture can endure these changes. This paper proposes a metric-based approach to evaluate architectural stability of a software system by combining these two traditional analysis techniques. Such an approach performs on the factbases extracted from the source code by reverse engineering techniques. We also present experimental results by applying the proposed approach to analyze the architectural stability across different versions of two spreadsheet systems.