My current float_sort correctly sorts positive floats with a comparable speedup to that for integer_sort. I'll deal with negative floats soon, but one issue popped up in testing that I wasn't expecting: nan's integer representation is a high positive number, yet with the float < operator it is treated as a very small (smallest?) number. Unless I special-case nan, the results of std::sort and float_sort will differ. Do I need to special-case nan? Also, there will only be a functor version of float_sort; I don't want to confuse the normal float >> (whatever that does) with the operation Spreadsort requires, and in my testing a good functor doesn't make a noticeable performance difference versus a member function. Given these issues, and the fact that the speedup is no greater than for integer_sort, is a float_sort still of interest? Steve