Take a look at this paragraph extracted from the description of FMULS in Atmel's AVR Instruction Set document (rev. G):
The multiplicand Rd and the multiplier Rr are two registers containing signed fractional numbers where the implicit radix point lies between bit 6 and bit 7. The 16-bit signed fractional product with the implicit radix point between bit 14 and bit 15 is placed in R1 (high byte) and R0 (low byte).
This confuses me slightly. Surely, for 8-bit signed fractional numbers, the radix point lies between bits 5 and 6 (bit 7 being the sign)? For 16-bit signed fractional numbers, the radix should lie between bits 13 and 14 (bit 15 being the sign). Am I wrong about this?