<p>Recently, the number of practical studies on two-­student pair learning in high school classes has been increasing. In such studies, researchers have encouraged students to share their roles, and to help their classmates by giving feedback to each other. However, studies have not examined the effect that pair learning has had on learning. They usually only examine formal pair learning, and they have not yet presented a methodology for pair learning that enhances the learning effect.</p><p>I reviewed the practices of, and research regarding, pair learning in Japan, and conducted a comparative study with peer learning studies in Europe and the United States. I then clarified issues in pair learning in Japan. Furthermore, I developed research methodologies to overcome these challenges.</p><p>In this paper, I have organized my findings into six points: a) the variables related to pair formation; b) the controlling of other variables, and calculating a constant; c) assigning non-­defined tasks such as essays, which require high levels of critical thinking; d) positioning formative assessment via general analytical rubrics; e) the necessities of structuring learning activities; and f) support for the argument that teachers should act as facilitators to help learners use the rubrics to eliminate their partners’ poor learning performance through feedback.</p><p>In this study, I investigated the learning effect that pair learning has on high school students in accordance with the six points explained above and compared it with my previous research on university students. The results clarified the following three points.</p><p>First, in the case of pair learning, which was evaluated based on rubric-­based formative assessment, it was shown that the effect on learning was large, with effect sizes of 0.78 and 0.91 obtained from high school students. These values were similar to the effect sizes produced by university students.</p><p>Second, the learning effects of pair learning have different meanings depending on the pair types. The results supported achievements made by university students through pair learning activities.</p><p>Third, it was found that gender differences were likely to impact the learning effect when pair learning was first performed. They also affected the pair types.</p><p>Furthermore, in my previous research, I was able to predict the pair types that appeared in pair learning activities with more than 75% accuracy based on a preliminary questionnaire, which contained 10 items on topics such as “collaboration” and “dominance,” whereas in this study, I were able to predict them with more than 80% accuracy.</p><p>This study covered SSH (Super Science High School) students, and the number of verified cases was small, so the results cannot be considered widely generalizable. However, they add to the research regarding the development future practice­-related research methodology.</p>