A method for generating atomistic models of dense amorphous polymers is presented. The generated models can be used as starting structures of Monte Carlo and molecular dynamics simulations, but also are suitable for the direct evaluation physical properties. The method is organized in a two-step procedure. First, structures are generated using an algorithm that minimizes the torsional strain. After this, an iterative algorithm is applied to relax the nonbonding interactions. In order to check the performance of the method we examined structure-dependent properties for three polymeric systems: polyethyelene (rho=0.85 g/cm(3)), poly(L,D-lactic) acid (rho=1.25 g/cm(3)), and polyglycolic acid (rho=1.50 g/cm(3)). The method successfully generated representative packings for such dense systems using minimum computational resources. (C) 2003 American Institute of Physics.