In this paper, a novel evolutionary programming with self-adaptive Cauchy mutation ACEP is proposed to solve the numerical optimization problems. ACEP utilizes the self-adaptive parameter r of Cauchy mutation to alter the search step size in time, and it obtains the best solution with only half population size of Fast evolutionary programming (FEP). The empirical experiments on four benchmark functions are undertaken. In the typical unimodal functions, ACEP performs much better than FEP on the convergence rate and the best solutions. While in the multimodal functions with many local minima, the accuracy of the best value in ACEP improves at least 50 percent than FEP on average statistic.