Modularization is a widely advocated mechanism to manage a business process model's size and complexity. However, the widespread use of subprocesses in models does not rest on solid evidence for its benefits to enhance their comprehension, nor are the criteria clear how to identify subprocesses. In this paper, we describe an empirical investigation to test the effectiveness of using subprocesses in real-life process models. Our results suggest that subprocesses may foster the understanding of a complex business process model by their "information hiding" quality. Furthermore, we explored different categories of criteria that can be used to automatically derive process fragments that seem suitable to capture as subprocesses. From this exploration, approaches that consider the connectedness of subprocesses seem most attractive to pursue. This insight can be used to develop tool support for the modularization of business process models.