Significance Determination of macromolecular crystal structures by molecular replacement (MR) uses a related structure to generate approximate phases for the unknown crystal. MR fails if the search model is too dissimilar to the unknown structure. In an accompanying paper, we described a method that generates many ensembles of possible solutions and refines them against the diffraction data without human intervention, and demonstrated its ability to solve test cases of known structure. Here, we apply this method to crystals of the epithelial cell polarity protein lethal giant larvae, which had resisted structure solution using conventional MR or experimental phasing. The results show that our method of unsupervised protein structure solution can be applied to a very large and difficult phasing problem.