When meeting the challenges in automatic and semi-automatic Web service composition, capturing the user's service demand and preferences is as important as knowing what the services can do. This paper discusses the idea of semantic service requests for composite services, and presents a way to model the elements of a composite service request as user preferences and constraints. The model is based on an interactive and iterative strategy meant to obtain the exact requirements from potential service consumers. The markup vocabularies and associated inference mechanism of OWL-S are used as a means to bring semantics to service requests. Language constructs are added to OWL-S as uniform representations of possible aspects of the requests. Using this model to represent the semantics of service requests enables the discovery agents to unambiguously understand the service need and precisely produce the desired composition. An application scenario is presented to illustrate how the proposed model can be applied to the real business world.