When using XSEM with either the STAX or SAX converters, dom4j is mangling certain XML documents as they are written into the index. Specifically, both converters extend AbstractXmlWriterXmlContentConverter, which in its toXml method uses dom4j's OutputFormat.createCompactFormat(). This output format is lossy - it trims whitespace from the document, so what you store into the index may not be what you get out again. This operation should be lossless. I've attached a test case illustrating the problem. The workaround is to create a custom subclass of the converter and override toXml() to use a lossless formatter.