That's still language design... Personally, I would make NaN == NaN in a language I would design, and provide a Float.equals function that would follow the IEEE standard.