Component-based software engineering has been projected off-late as an appropriate paradigm for embedded systems development. However this becomes feasible only if the dependability attributes, such as reliability, of embedded systems can be properly estimated and guaranteed using appropriate models. Another recent trend is the prevalence of Architectural modeling of software systems due to the advantages such as composability reasoning and analysis of system behavior as well as the flexibility and reusability of models. However, there is little support for the systematic modeling of reliability in architectural description languages at a given abstraction level. Another major issue is the uncertainty of the reliability parameters associated with the traditional testing-based approaches for particular architectural entities (e.g. components and/or connectors). In this context, we propose a new approach for the reliability estimation of component-based software systems, which steps on an architectural model of the system using Generalized Nets. This notation is a generalization of Petri nets and is an all-purpose formal modeling technique for the description of parallel phenomena. For the reliability description, we apply possibility theory, which has the potential to solve the uncertainty aspect. This is based on fuzzy sets and describes reliability properties of software components in systems architecture with fuzzy measures.