This article proposes an automatic genetic algorithm in clustering for discrete elements (AGAD). In this algorithm, the suitable number of clusters is determined by a new measure called the similar index of cluster and the steps of traditional genetic algorithm are improved. The convergence of AGAD is considered by theory and illustrated by numerical examples. The proposed algorithm is tested by many numerical examples and performed by Matlab procedure. These examples illustrate the superiority of the proposed algorithm over some existing algorithms. They also demonstrate the feasibility and applicability of studied problem.