Due to the increasingly competitive global market environment and according to the share of costs of delivery and product distribution in the entire costs of supply chain. Designing an appropriate and efficient distribution network is very important issues in distribution network design can be referred to include determining the number, location and capacity of warehouses and factories, the optimum strategy to transfer product from the factory to warehouse and customer and from warehouse to customer. In this thesis, problem of designing the distribution network in a multilevel supply chain has been studied considering the possibility of direct transferring goods from production equipment to consumer. The possibility of direct transfer from Plant to consumer is raised due to today's the make to order strategy is highly regarded.The problem modeled as mixed integer programming in three cases: a single product, Multi-Product and single product with limited distribution radius. The first model is static and definitive. Any production and distribution equipment have multi-level capacity and the number of them in the model has been considered variables. Second model is written as the first model development with conditions and characteristics considered in the first model and to transfer of some product in distribution network. Since there is probably that a company can't transfer product to all distribution centers or customers and also there is probably a distribution center can't transfer product to all customers, so, in the third model has been considered restriction of distribution radius for production plant and distribution centers (warehouse). Despite this limitation, if distance between a plant and a distribution center or a customer to be less than Plant radius, then plant can transfer the product to them. Also, if distance between a distribution center and a customer to be less than distribution radius, then a distribution center can transfer the product to customer. Since the issue of designing a distribution network is NP-hard problem, so a Meta-heuristic method as particle swarm optimization (PSO) method has been used. To evaluate the effectiveness of the presented approach has been solved 16 issue of first model, 32 issue of II Model and 18 issue of the III model and results of solving them using the proposed PSO algorithm has been compared with the results of solving them by Lingo software. Comparing the results indicate efficacy of presented Method.