Many protein structure comparison methods use secondary structure information to do fast structure similarity search for initial alignment finding and refine the results from possible optimal candidate solutions by iteratively dynamic programming to optimize the final results. In this paper, we develop a method, Ellipsoidal Model Protein Structure Comparison, based on the concept of secondary structure elements alignment followed by iteratively refinement. In order to utilize all possible structure information to obtain alternative solutions for further analysis, we use ellipsoidal model to represent not only mainly -helices and -sheets, but the remaining fragments for structural alignment. Different heuristic filters and geometric hashing based global alignment estimation are applied for quick finding better initial alignments. We also provide top-N solutions without increasing extra computational time rather than only best solution in the previous works. Now, we provide the online web service, Ballerina (http://ballerina.csie.ntu.edu.tw/), for protein structure comparison.