In the past 30 years, although much effort has been made to narrow the gender gap in science, technology, engineering and mathematics (STEM), females are still largely underrepresented in some important STEM fields, such as physics and engineering (NSF, 2007). To deal with this situation, people from different sectors have long reached a common understanding: Educators must improve school girls’ interest, participation and engagement in STEM subjects (e.g., Office of Science and Technology Policy, 2013). In the K-12 classroom, small group work has been shown to promote an equitable environment for girls’ learning in science and have a positive impact on their persistence in STEM disciplines (e.g., Davis & Rosser, 1996). Further research shows that same-gender grouping enhances girls’ engagement and achievement in STEM fields (e.g., Riordan, 1990). However, little research has been done in design-based science (DBS), a pedagogy that allows students to learn science through engineering design, which is considered as important as inquiry-based learning (NGSS, 2013). This study was an effort to make contributions in this aspect. In two DBS tasks in high school biology, this study arranged various small group gender compositions: from 33% to 100% female. In these contexts, this study explored (1) How gender composition influenced girls’ and boys’ engagement; (2) how student engagement influenced their achievement, and (3) how group gender composition influenced girls’ and boys’ achievement in engineering practices and biology content. Results show that higher group female percent led to higher engagement levels and engineering practice achievement of girls. However, group cohesion and positive group interaction were indispensable as they were needed for girls (and boys, in certain cases) to develop senses of relatedness and collective efficacy, which were necessary for their engagement and learning. Also, results show that group gender composition wasn’t only directly correlated with girls’ achievement, but also indirectly correlated with this variable through the mediation of the girls’ behavioral, emotional and cognitive engagement, respectively. Based on these findings, implications for classroom teaching and future research are provided.