With the introduction of the XML data type (using an XML data model) comes there's a choice of pulling XML out of the database as a stream or column. Using an XML data type as a column type will get you a column of type XML. That's SqlXml to you in .NET. SELECT … FOR XML gets you a stream. In SQL Server 2005 there a new keyword TYPE on SELECT … FOR XML that makes the stream into an XML type. Some/most tools make the stream look like a column with up to a size of 4k per "row". SSMS in SQL Server 2005 does better with it than Query Analyzer.
There's different APIs for these too. In ADO.NET there is ExecuteXmlReader vs ExecuteReader/ExecuteScalar. In OLE DB you use a COM IStream implementation, there was a Stream class added to ADO classic. Although you can use the either API for stream or column, it appears to be a stretch to use “other one“ in both cases. As an example, using a 'SELECT xmlcol from xmltab' with ExecuteXmlReader only gets you the XML in the first row. As you'd expect, if you remember ExecuteXmlReader returns a singleton like ExecuteRow in ADO.NET 2.0 does. At this point, I'm recommending using the "right" API for the representation. And you can always translate to a string. And, using the client XML APIs, any of the supported XML data models (such as document or Infoset).