Abstract: An efficient differential evolution algorithm (AEDE) for function optimization is proposed. First, with population evolution, AEDE divides population into three groups by the fitness’s normal distribution and the three groups adopt different mutation operators. Second, the selection of the individuals involved in mutation operation uses alternatively a random method and a roulette wheel method based on affinity matrix. To validate the superiority of AEDE, AEDE and some state-of-the-art DE variants proposed in pertinent literatures are compared as regards nine benchmark functions. The simulation results show that ANDE promises competitive performance not only in the convergence speed but also in the quality of solution.