In service-oriented computing, services are dynamically built as an assembly of pre-existing, independently developed, network accessible services. Hence, predicting as much as possible automatically their dependability is important to appropriately drive the selection and assembly of services, in order to get some required dependability level. We present an approach to the reliability prediction of such services, based on the partial information published with each service, and that lends itself to automatization. The proposed methodology exploits ideas from the Software Architecture- and Component-based approaches to software design.