Conceptual modeling is an important part of Enterprise Modeling, which is a challenging field for both teachers and learners. Creating conceptual models is a so-called ‘ill-structured’ task, i.e. multiple good solutions are possible, and thus students can follow very distinct modeling processes to achieve successful learning outcomes. Nevertheless, it is possible that some principles of modeling behavior are more typical for high-performing rather than low-performing students, and vice versa. In this study, we aimed to discover those patterns by analyzing logged student modeling behavior with process mining, a set of tools for dealing with event-based data. We analyzed data from two individual conceptual modeling assignments in the JMermaid modeling environment based on the MERODE method. The study identified the presence of behavioral patterns in the modeling process that are indicative for better/worse learning outcomes, and showed what these patterns are. Another important finding is that students’ performance in intermediate assignments is as well indicative of their performance in the whole course. Thus, predicting these problems as early as possible can help teachers to support students and change their final outcomes to better ones.