Software has been and is still mostly refactored without tool support. Moreover, as we found in our case studies, programmers tend not to document these changes as refactorings, or even worse label changes as refactorings, although they are not. In this paper we present a technique to detect changes that are likely to be refactorings and rank them according to the likelihood. The evaluation shows that the method has both a high recall and a high precision ? it finds most of the refactorings, and most of the found refactoring candidates are really refactorings.