The solution to multi-objective optimization problems with conflicting objectives is a Pareto-optimal solution set. It is well known that the critical work in multi-objective particle swarm optimization (MOPSO) is to find the global best guides for each particle in order to obtain satisfied Pareto fronts with high diversity. In this paper, a modified version of MOPSO is proposed, where dense and sparse distance are adopted to determine the global best guides, and Pareto archive with size limit is used to store the non-dominated solutions. In addition, a random number is used to judge whether the crowding distance considered or not, and the inertia weight decreases linearly to improve the speed of convergence and avoid precocity. The proposed approach is applied to several well-known benchmark functions, and the experimental results show that the diversity of swarm and distribution of Pareto fronts are well satisfied.