We present two fundamental results about the computational power of affine finite automata (AfAs). First, we show that changing the cutpoints does not change the class of languages defined by AfAs. Second, we show that AfAs define more languages than probabilistic and quantum finite automata with exclusive cutpoints.