Abstract: A new algorithm for getting core attributes is firstly given in this paper.Then simulated annealing genetic algorithm is improved by dynamic regulation of the nearest neighbor subset and used in minimal attribute reduction of decision table.This algorithm uses core attributes to optimize the initial population and adaptive crossover probability and mutation probability，restrains premature convergence effectively and enhancs the ability of search and efficiency.The experiment results demonstrate that the algorithm is efficient and feasible in minimal attribute reduction.