A class of goodness-of-fit statistics including that of Freeman and Tukey for testing discrete models is introduced; optimum statistics are identified through comparison of asymptotic moments with those of power divergence statistics and via simulations. Further new classes of test statistics are indicated.