As the Web Services and Grid community adopt Semantic Web technology, we observe a shift towards higherlevel workflow composition and service discovery practices. While this provides excellent functionality to non-expert users, more sophisticated middleware is required to hide the details of service invocation and service integration. An investigation of a common Bioinformatics use case reveals that the execution of high-level workflow designs requires additional processing to harmonise syntactically incompatible service interfaces. In this paper, we present an architecture to support the automatic reconciliation of data formats in such Web Service worklflows. The mediation of data is driven by ontologies that encapsulate the information contained in heterogeneous data structures supplying a common, conceptual data representation. Data conversion is carried out by a Configurable Mediator component, consuming mappings between XML schemas and OWL ontologies. We describe our system and give examples of our mapping language against the background of a Bioinformatics use case.