<p>Computational thinking (CT) is a core skill for students in the digital age, which refers to the use of fundamental concepts of computer science to analyze and solve real-world problems. However, it is a challenging task to accurately measure the CT levels of the students. The goal of this study is to explore a new CT assessment paradigm that combines a data-driven approach with evidence-based reasoning, and to verify the reliability and validity of this paradigm through an empirical study. Specifically, we examined middle school students’ CT by adopting an Evidence-Centered Design (ECD) approach. We first developed a CT evaluation framework based on ECD which included student, task, and evidence models. We then created several simulation-based tasks to gather evidence of CT levels; with these tasks, we extracted the observable variables in ECD model from clickstream data during the students’ answer process. Finally, a Bayesian network model was used for CT level prediction. Findings suggest that students’ clickstream data provide fine-grained, time-variant information on their interactions with tasks, thus promising more objective and richer insight into the performances of students with different CT levels. This study provides a paradigm reference for future CT measurement studies and proposes ideas for using process data in teaching to promote students’ CT development.</p>