This study examines the role of self-regulated learning (SRL) in facilitating students' shifts to more sophisticated mental models of the circulatory system as indicated by both performance and process data. We began with Winne and colleagues' information processing model of SRL (Winne, 2001; Winne & Hadwin, 1998) and used it to examine how students regulated their own learning when using a hypermedia environment to learn about the circulatory system. Undergraduate students (N = 24) were trained to use a hypermedia environment to learn about the circulatory system. Pretest, posttest, and verbal protocol data were collected to measure the shifts in conceptual understanding (from pretest to posttest) and the SRL variables associated with shifts in conceptual understanding. We used a median split to divide the sample into two groups of learners--high-jumpers and low-jumpers (i.e., students who either showed large gains in conceptual understanding or showed relatively little or no gain in their conceptual understanding). Findings revealed that the high-jumpers shifted an average of 4.5 mental models, while the low-jumpers had an average shift of less than one mental model from pretest to posttest. The verbal protocol data were coded for various self-regulating variables to examine which SRL variables differentiated the low-jumpers from the high-jumpers. High-jumpers were much better at regulating their learning during the knowledge construction activity. In general, they used effective strategies, planned their learning by creating subgoals and activating prior knowledge, monitored their emerging understanding, and planned their time and effort. In contrast, low-jumpers used equal amounts of effective and ineffective strategies, planned their learning by using sub-goals and recycling goals in working memory, handled task difficulties and demands by engaging mainly in help-seeking behavior, and did not spend much time monitoring their learning. Our results provide a valuable initial characterization of the role of SRL in accounting for differences in conceptual knowledge gains when students use hypermedia environments to learn about complex science topics. In addition, we discuss the general instructional implications of our findings and how they can be used to inform the design of hypermedia environments to teach students about complex systems.