lexical_cast does not allow to pass the locale to be used during conversion. This is justified with analogy of lexical_cast to other casts which all take only one parameter. So lexical_cast always uses the current global locale. A reasonable choice. However I wanted to propose some additional features. It would be useful to provide (lets call it so for now) classic_lexical_cast which would use std::locale::classic() instead of the global one. I just finalized globalization/localization of our application by adding setting of the global locale to appropriate value according to the selected language. I noticed that all of sudden serialized data (mainly projects) stopped loading/saving correctly. The reason was easy to find. Where previously we had XML node named "item1000" we now have "item1,000" in English or "item1 000" in Polish for example. The projects are no longer interchangeable between applications running in different language version. I corrected this by adding our own lexical_cast which resets the global locale to the classic one, calls boost::lexical_cast and finally resets the global locale back. This special cast is used now whenever we are dealing with texts which are not presented to the final user. This works fine however it seems that resetting the global locale often (each time we save a number in the project for example...) is causing significant performance overhead. (Exact tests have not been yet performed but this seems the only change which might have affected all serialization slowdown.) I might reset it before saving the project and then save the project as usually but this is not very handy as I would have to remember to do it each time I am serializing any data. Also other problems are implied. Other threads for the short period of time see different global locale as well and thus might produce bad strings. And in the end it is impractical. It would be handy to have classic_lexical_cast as well as it could introduce some optimizations knowing that the requested locale is the classic one. Another thing is that it would be handy to introduce a function textual_convert (another example name) which would act like lexical_cast but would take the requested locale as a parameter (this is no longer a cast so it may). lexical_cast might take that locale as a second parameter with default value being the global locale so it could still be used just as any other casting. But I suppose this would not be well seen. Thus why not introduce a different function for that (or maybe there already is one?). It seems those changes would not be difficult to implement and could be hidden in the details of lexical_cast easily. What do you think of that? Finally lexical_cast's documentation might mention explicitly that current global locale is used as it does not mention that if I am not mistaken (I haven't found "locale" word in that document). Adam Badura