As highlighted by the World Wide Web Consortium, XML has been proposed to deal with huge volumes of electronic documents and is playing an increasing important role in the exchange of a wide variety of data on the Web. However, when dealing with such large and heterogeneous data sources, it is necessary to have an idea on the way these data sources are structured. This information is indeed essential in order to build mediator schemas. These mediator schemas are required to query data in a uniform way. Moreover, this information is interesting since it provides users with a semantic structure of the data they can query. Recently schema mining approaches have been proposed to extract in an efficient way the commonly occurring schemas from a collection. Nevertheless, according to the semantic point of view, such approaches suffer from different drawbacks. In this work, we propose thus a fuzzy approach, showing why and how fuzziness is useful in order to extract frequent approximate schemas.