The Shift-and-Mask approach outlined by Per is completely portable and completely general - it can be adapted for extracting any sized portions out of any (larger) sized word on any platform.

Depending on the compiler, it may not even be any slower than the union...


Here it is again:

u24_t var24; // Some assumed 24-bit (3-byte) tytpe

//                      Shift as required  Mask-off the required portion
//                      -----------------  -----------------------------
unsigned char low    =  var24              & 0xff;
unsigned char middle = (var24 >>  8)       & 0xff;
unsigned char high   = (var24 >> 16)       & 0xff;