Abstract: We propose a saddlepoint approximation of the error probability of a binary hypothesis test between two i.i.d. distributions. The approximation is accurate ...