We present a de novo model-building approach that combines predicted backbone conformations with side-chain fit to density to accurately assign sequence into density maps. This method yielded accurate models for six of nine experimental maps at 3.3- to 4.8