The paper presents a new automated method for modeling loops in protein structures, which significantly improves the accuracy of loop predictions. The method optimizes the positions of all nonhydrogen atoms of the loop in a fixed environment using a pseudo energy function that includes terms from the CHARMM-22 force field, statistical preferences for dihedral angles, and nonbonded atomic contacts. The energy function is optimized using conjugate gradients, molecular dynamics, and simulated annealing. The predicted loop conformation corresponds to the lowest energy conformation among 500 independent optimizations. The accuracy of loop predictions was evaluated for 40 loops of known structure at each length from 1 to 14 residues. When accuracy is measured by local superposition of the model on the native loop, 100, 90, and 30% of 4-, 8-, and 12-residue loop predictions, respectively, had RMSD error less than 2 Å for the main-chain N, Cα, C, and O atoms. The average accuracies were 0.59±0.05, 1.16±0.10, and 2.61±0.16 Å, respectively. The accuracy of the lowest energy prediction for a given loop can be estimated from the structural variability among a number of low energy predictions. The method was compared with one of the most successful previously described methods and its accuracy in recent blind predictions of protein structure. The accuracy of the method is primarily limited by the accuracy of the energy function rather than the extent of conformational sampling. The method was tested on 14 test sets of loops, each containing 40 loops of the same length, spanning the range from 1 to 14 residues. The accuracy of loop predictions was evaluated as a function of thoroughness of conformational sampling, loop length, and structural properties of native loops. The results show that the accuracy of loop predictions decreases with increasing loop length. The method was also evaluated in distorted environments, where the accuracy of loop predictions was found to be worse than in the native environment. The method was found to be more accurate than previous methods, particularly for longer loops. The method is efficient and robust, with predictions for many loops under various conditions performed in parallel on a cluster of SGI and PC Linux computers. The results demonstrate that the method is a significant improvement over previous methods for modeling loops in protein structures.The paper presents a new automated method for modeling loops in protein structures, which significantly improves the accuracy of loop predictions. The method optimizes the positions of all nonhydrogen atoms of the loop in a fixed environment using a pseudo energy function that includes terms from the CHARMM-22 force field, statistical preferences for dihedral angles, and nonbonded atomic contacts. The energy function is optimized using conjugate gradients, molecular dynamics, and simulated annealing. The predicted loop conformation corresponds to the lowest energy conformation among 500 independent optimizations. The accuracy of loop predictions was evaluated for 40 loops of known structure at each length from 1 to 14 residues. When accuracy is measured by local superposition of the model on the native loop, 100, 90, and 30% of 4-, 8-, and 12-residue loop predictions, respectively, had RMSD error less than 2 Å for the main-chain N, Cα, C, and O atoms. The average accuracies were 0.59±0.05, 1.16±0.10, and 2.61±0.16 Å, respectively. The accuracy of the lowest energy prediction for a given loop can be estimated from the structural variability among a number of low energy predictions. The method was compared with one of the most successful previously described methods and its accuracy in recent blind predictions of protein structure. The accuracy of the method is primarily limited by the accuracy of the energy function rather than the extent of conformational sampling. The method was tested on 14 test sets of loops, each containing 40 loops of the same length, spanning the range from 1 to 14 residues. The accuracy of loop predictions was evaluated as a function of thoroughness of conformational sampling, loop length, and structural properties of native loops. The results show that the accuracy of loop predictions decreases with increasing loop length. The method was also evaluated in distorted environments, where the accuracy of loop predictions was found to be worse than in the native environment. The method was found to be more accurate than previous methods, particularly for longer loops. The method is efficient and robust, with predictions for many loops under various conditions performed in parallel on a cluster of SGI and PC Linux computers. The results demonstrate that the method is a significant improvement over previous methods for modeling loops in protein structures.