The use of sub-nanometer resolution electron density as spatial constraints for de novo and ab initio structure prediction requires knowledge of protein boundaries to accurately segment the electron density for the prediction algorithms. Here we present a procedure where even poorly segmented density can be used to determine the fold of the protein. The method is automated, fast, capable of searching for multiple copies of a protein fold, and accessible to densities encompassing more than a thousand residues. The automation is particularly powerful as it allows the procedure to take full advantage of the expanding repository in the Protein Data Bank. We have tested the method on nine segmented sub-nanometer image reconstruction electron densities. The method successfully identifies the correct fold for the six densities for which an atomic structure is known, identifies a fold that agrees with prior structural data, a fold that agrees with predictions from the Fold & Function Assignment server, and a fold that correlates with secondary structure prediction. The identified folds in the last three examples can be used as templates for comparative modeling of the bacteriophage P22 tail-machine (a 3MDa complex composed of 39 protein subunits).