The fill method on mb_from_wchar calls std::wctomb, which is backed by 'WideCharToMultiByte'. This function uses the codepage defined in the current locale, however, the utf-8 codepage does not function properly on VC++ 7 or 8. Would it be difficult to have the conversion use the code_cvt facet defined for the stream underlying the archive? It also seems that doing so would ensure serialization / deserialization would be more consistent and self-sufficient, since neither process would depend on a (mutable) global locale setting. At any rate, the serialization library works with certain window's codepages, but *not* with the utf-8 codepage (65001). It would be nice to have the library use its own utf8_code_cvt facet to translate wide characters to multibyte. If there is some work around I'm not aware of, I would appreciate any advice. Thanks.