Conditional optimization allows the incorporation of extensive geometrical information in protein structure refinement, without the requirement of an explicit chemical assignment of the individual atoms. Here, a mean-force potential for the conditional optimization of protein structures is presented that expresses knowledge of common protein conformations in terms of interatomic distances, torsion angles and numbers of neighbouring atoms. information is included for protein fragments up to several residues long in alpha-helical, beta-strand and loop conformations, comprising the main chain and side chains up to the gamma position in three distinct rotamers. Using this parameter set, conditional optimization of three small protein structures against 2.0 A observed diffraction data shows a large radius of convergence, validating the presented force field and illustrating the feasibility of the approach. The generally applicable force field allows the development of novel phase-improvement procedures using the conditional optimization technique.