Background: Scholars have confirmed the vital roles of self-regulated learning (SRL) behaviours in predicting task performance, especially within non-linear technology-rich learning environments (TREs). However, few studies focused on the learning costs (e.g., study effort and time-on-task) related to SRL and the efficiency outcome of SRL (i.e., the relative relationship between learning costs and performance). Objectives: This study examined the relationship between students' SRL behaviours and problem-solving efficiency in the context of TREs. Methods: Eighty-two medical students accomplished a diagnostic task in a computer-simulated environment, and they were classified into the efficient or less efficient group according to diagnostic performance and time-on-task. Then we coded students' SRL behaviours from trace data and counted the frequency of each SRL behaviour. The recurrence quantification and lag sequential analyses were performed to extract the dynamic characteristics of SRL behaviours, including recurrent patterns and sequential transitions. Results and Conclusions: Efficient students conducted more frequent Self-reflection behaviours than the less efficient. For the recurrent patterns, efficient students tended to exhibit longer SRL behaviour sequences comprising a variety of different SRL behaviours (e.g., Task Analysis > Add Test > Add Hypotheses > Categorise Evidence) as well as longer sequences of repeated SRL behaviours (e.g., Add Test > Add Test > Add Test > Add Test). Moreover, efficient students exhibited more sequential transitions between different SRL behaviours than less efficient. Takeaways: Overall, this study revealed the effects of SRL on problem-solving efficiency, which inspired researchers to incorporate problem-solving efficiency as an evaluation criterion of SRL processes.