The binary improved particle swarm optimization(PSO) algorithm for knapsack problem is brought forward,and the detailed realization of the algorithm is illustrated.In order to speed up the convergence,the memory mechanism is implanted in the traditional binary PSO.Some examples in other references are recomputed and both results are compared.It can be found that the algorithm presented is better than genetic algorithm and simulated annealing algorithm in the ability of finding optimal value,the speed and the computation stability.The algorithm proposed can be applied to other discrete optimization problems.