The development of an integrated software environment for protein structure refinement is reported. Energy minimization is combined with geometric embedding in the refinement program. The energy minimization procedure is used to sample the conformational space and find a group of low energy structures for further improvement. The distance geometry also known as geometric embedding is then applied to the structures with a set of statistical distances (distance derived statistically from known protein structures). The CHARMM potentials along with a set of recently developed statistical potentials are used in energy minimization. For distance geometry, in addition to the statistical distances, a set of distance bounds is also generated for each of the structures based on their normal mode fluctuations. The final output of the refinement program is an ensemble of plausible structures. The implementation of the algorithms, the organization of the software, and the parallelization of the computation is described. Some sample refinement results are also presented.