I don't know that I can offer you a good solution. But here is what I can come up with. First a couple of relevant facts about the serialization library. a) generally it saves in the archive only that information absolutely required to be able to recover the data. This is basad on design/efficiency considerations. b) In some cases it stores "optional" information. That it is, the archive may be written to save and recover this information. Normally these are just stubbed out as no ops. c) In xml archives, information about object and class id and their references is included in the xml. I'm don't remember whether its done in the "optional" types or not. c) class name was not considered as one of these optional types i) it wasn't deemed necessary ii) it is not always available. Many types (e.g. primitives) are serialized without calling the library code. For these cases the "optional" data is not called. So back to your situation. What is really needed for you is the following: a) some sort reflection for C++ so that you can write something like t.name(). b) derive from xml_oarchive to invoke t.name() and insert it into the xml stream output. Examples of deriving from one archive type to produce another are in the documentation. So the problem is "reduced" to making/finding something like t.name() which returns a text string which can be inserted into the output stream. This already exists on one form - the RTTI implementation uses type_id(t).name to return a pointer to a name. This may be sufficient for your needs. The only problem with it is that a) the name will be mangled to a unique C++ name. This may not always be user friendly. b) the name will be mangled differently for each compiler. If you can work with the above limitations - perhaps by massaging the mangled name to something readable - this can probably be made to work for you. If not - you can use the name passed by BOOST_CLASS_EXPORT which was "invented" to address the issues above. The only problem is that BOOST_CLASS_EXPORT might not bedefined for all your types. You should have enough here to craft a solution to your problem. Good Luck Robert Ramey "Eduardo Ferro Aldama" <eduardo_ferro@carrefour.com> wrote in message news:OFCDA9E07E.A71F2EB0-ONC12571AA.003AE8AF-C12571AA.003AE8B4@carrefour.com... Hi! We are trying the boost's serialization-lib to serialize C++ objets to send them throw the network. Initialy, we are use xml-serialization to facilitate the inspect the comunications. We are trying to develop a little tool to work/inspect this messages (xml-serializated) but we need to develop this tool in another languaje (in this case python) and We want that this tool don't need to know nothing about the original C++ clases.... In order to develop this tool we need add to the output xml string information about the names of the class that we are serializing -Is possible to configure something to add the class name in the xml output format? probably -Can we add any other information to taht format without disturb the deserialization process?? you have to add it the ouput - and skip in on the input. xml input archives already skip irrelevent attributes so this might not be a problem. Some hint/tipo/doc to add this functionality without changing the boost-lib source? se above. Thanks in advance you're welcome in advanced Please accept my apologies for my poor english no problem - I live in california.