To solve the order task allocation in Dynamic Virtual Enterprise (DVE), a multi-goal decision-making model was constructed. As to the model, an improved Particle Swarm Optimization (PSO) algorithm based order task allocation method was presented on the basis of analyzing the traditional PSO algorithm. The method overcome the deficiencies of prematurely and trapped into local optimal solution easily in traditional PSO by adjusting the parameters in algorithm automatically and introducing mutation operation from genetic algorithm. And the TOPSIS based computing method of particle fitness value was researched. Finally, the feasibility of the method was verified by an application example.