Recent years have witnessed a growing interest in using Web Services as a powerful means for data publishing and sharing on top of the Web. This class of services is commonly known as DaaS (Data-as-a-Service), or also data services. The data returned by a data service is often subject to uncertainty for various reasons (e.g., privacy constraints, unreliable data collection instruments, etc. In this paper, we revisit the basic activities related to (Web) data services that are impacted by uncertainty, including the service description, invocation and composition. We propose a probabilistic approach to deal with uncertainty in all of these activities.