The forming process of powder bed for additive manufacturing (AM) is analyzed and is simplified to three processes, including random packing, layering, and compression. The processes are simulated by using the discrete element method (DEM). First, the particles with monosize, bimodal, and Gaussian size distributions are randomly packed. Then, the packed particles are layered with different thicknesses. Finally, a 20 μm compression is applied on the top surface of the layered powder beds. All the processes are simulated based on the soft sphere model. Packing density and coordination number are calculated to evaluate the packing mesostructure. The results indicate that the packing density and coordination number increase with the layer thickness increasing in the initial packing, and compression can effectively increase the density and coordination number of powder bed and decrease the effect of ranging layer thickness. The results also show that powder bed with monosize distribution initially has the best combination performance. Our research provides a theoretical guide to choosing the layer thickness and size distribution initially of powder bed for AM.