>Not if it's a signalling NaN. If the chip only supports QNaNs, then >you might be able to get away with it. Flipping the sign bit is correct even for a signalling NaN. See item 2 in the Appendix of IEEE Std 754-1985 (Binary Floating-Point Arithmetic). See also item 2 in the Appendix of IEEE Std 854-1987 (Radix-Independent Floating-Point Arithmetic). Jim