It seems that most xml is in UTF-8. However, whenever I have a XmlTextWriter create xml in UTF-8 it always puts three additional bytes at the start that screws things up. The resulting xml will not load into a XPathDocument or XmlDocument. If I use ascii encoding it works fine. Can anyone tell me what's happening here?

those 3 bytes is the Byte Order Mark added because of the Encoding. Don't know total definition for the Byte Order Mark but it's kind of a signature with encoding info and byte order.

You can use another stream with an offset of 3 to remove the extra bytes... or you can change a little your code and even make it more simple... for instance use StringWriter instead of MemoryStream, or create a new class that is derivated from StringWriter where you can pass the Encoding you want, this way:

private class MyStringWriter : StringWriter
{
    Encoding encoding;

    public MyStringWriter(Encoding encoding)
    {
        this.encoding = encoding;
    }

    public override Encoding Encoding
    {
        get {
                     return encoding;
                }
    }
}

your method becomes like this:

  MyStringWriter sw = null;
  XmlTextWriter xtw = null;
  string ret = "";

  sw = new StringWriterWithEncoding(Encoding.UTF8);
  xtw = new XmlTextWriter(sw);
  xtw.WriteStartDocument();
  xtw.WriteStartElement("Import");
  xtw.WriteStartElement("Indexes");

  xtw.WriteStartElement("Index");
  xtw.WriteAttributeString("name", "ControlNumber");
  xtw.WriteAttributeString("value", "123456789");
  xtw.WriteEndElement();

  xtw.WriteEndElement();
  xtw.WriteEndElement();
  xtw.WriteEndDocument();
  xtw.Flush();

  ret = sw.ToString();