Due to the increasingly widespread use of XML, many XML-related applications require the service of schema mediation, which is to find semantically similar elements from two or more schema sources. Current approaches to schema mediation require much improvement. First, they do not cope well with \OUT{handling} a large number of XML schemas since most involve repeating pairwise integration. Second, they do not provide sufficiently flexible solution because most mediated schemas commit to some particular choice regarding the representation of relationships between elements. This paper proposes a novel schema mediation approach, called {\em XMediator}, to address these problems. It exploits the structural properties of XML schemas, and generates a mediated schema from multiple schema sources\OUT{which can materialized into multiple tree structures that can be used in a highly flexible manner for applications that use documents -- needs to define a measure like 'coverage' or tree-subsumption}. Experimental evaluations on both synthetic and real datasets show the effectiveness of {\em XMediator}.