First thing I'd say is don't be confused by "Big Endian" on ARM. There aren't 2 endian-nesses available with ARM, there's 4: 1. little endian with devices connected little endian 2. little endian with devices connected big endian 3. big endian with devices connected little endian 4. big endian with devices connected big endian Only (1) and (4) are true little/big endian. (2) and (3) are fake. Each one of these needs handling, since I believe most people refer to (3) as big endian, which is, according to the ARM manuals, incorrect. In other words, you can't just take as assabet, switch the endian bit in the control register and say "I have a big endian machine". Why do these have to be handled differently? Each one affects the way words and bytes are accessed in peripherals, and may or may not result in a greater amount of code to sort out the endian issues. It is debatable whether the (3) is really worth the effort due to the code that gets added into critical functions. Ok, here's why. The following describe the layout of bytes that ARM processors expect on the data bus. A word is always D31-D0 = bit 31 - bit 0.